Exclusive: anonymous flagging service says cases have surged, as watchdog says AI is making sexualised or ‘nudified’ content easier to produce
Children in the UK have reported a surge in explicit deepfake images featuring their likenesses, as a safety watchdog warned AI is making such content easier to produce.
The Report Remove service, which blocks intimate images from appearing online, said there had been a rapid increase in imagery that had been digitally manipulated or “faked” in some way, including with AI and nudification tools.
One summer day earlier this year, Rhys left a cafe in Melbourne where he had been working remotely, when his phone pinged with a message on the dating app Grindr.
The message was a photo of him sitting at a table outside the cafe on his laptop, with his dog seated on the ground next to him. In the foreground, in a man’s hand was an iPhone unlocked.
Up to 1,500 items were identified as missing or stolen from the British Museum following a review of its collection in 2023.Flik47/Shutterstock
In the last few years, some of the UK’s biggest cultural institutions have experienced major threats to their security. A report by the public accounts committee (PAC) has warned that such incidents have exposed serious weaknesses across the sector but that the government has failed to devise a strategy to prevent them. This has left Britain’s cultural institutions open to similar future attacks.
We live in a time when digital systems underpin everything from ticketing to surveillance and building access. Sidelining cybersecurity leaves museums vulnerable to attacks that could threaten both their finances and their collections.
If the UK wants to avoid major attacks and thefts, it needs to learn a lesson from the recent Louvre heist where €88 million (£78 million) of jewels were stolen. An audit conducted just weeks before the incident exposed the Louvre’s spending priorities that left them vulnerable to theft. The report found that €169 million was spent on artwork and exhibitions but only €26.7 million on all forms of maintenance, including security.
The likely thinking of cultural institutions is that to remain financially resilient they have to spend more on money-making attractions, like artwork and exhibitions. However, this means that security sometimes becomes a low priority – less money is spent and less thought given to the importance of keeping it up to date. However, the financial loss from theft affects financial resilience by eroding emergency cash reserves and driving up insurance premiums.
As the Louvre’s audit shows, the museum could have perhaps avoided the theft had it taken its findings more seriously.
The issue of outdated security measures at the Louvre goes back to at least 2017. An audit at the time found that “certain workstations [had] obsolete operating systems (Windows 2000 and Windows XP) which no longer guarantee effective security”. Security updates for Windows 2000 ceased in 2010, and Windows XP in 2014.
Three weeks after the heist in 2025, the Louvre was criticised by France’s Court of Auditors for not taking the audits seriously and spending money on art instead of security in the years before the theft.
Similar weaknesses in investment priorities as the Louvre’s were exposed in a report made by The British Library in March 2024 into their cyber-attack.
In October 2023, The British Library experienced a severe cyber-attack. The library’s systems were infiltrated by hackers who locked out all network users and demanded a ransom of 20 Bitcoin (approximately £590,000). The ransom was not paid and the stolen data was auctioned and then leaked on the dark web.
The library’s report into the attack acknowledged that it happened because it had muddled through with a historical mixture of old systems from many sources, which didn’t have a recovery plan. This meant it took the British Library months to restore the most basic tool – the online catalogue. And even now, five major services, including the catalogue of illuminated manuscripts, remain unavailable.
The cultural institutions, government and even the PAC make the classic mistake of divorcing cybersecurity from physical security.
The report by PAC does mention “bringing together chief digital information officers and chief information security officers” and separately it praises the “National Museum Security Group”. However, there is no mention of encouraging dialogue between the groups.
This divide is very common, and is often built into organisational system. The most junior person responsible for both physical and cybersecurity is often the CEO, or possibly a chief operating officer (COO). This structure worked in the days when physical security did not rely on computers. However, today, when so much physical security relies upon technology, it just does not make sense. There should be someone further down the chain, such as a “chief security manager” or equivalent.
As the report by Apolo Security into the Louvre attack found: “this case highlights another growing challenge: the convergence between operational technology (OT) and information technology (IT). When camera, climate or access control systems are connected to the network, any digital divide can have immediate physical consequences.”
This false split between standard IT cybersecurity (the domain of the chief information security officer), operational technology (the unnoticed computer-enabled devices that keep things working like electronic doors or intruder alarms) and physical security is far from unique to the scope of PAC’s report.
A classic example is the cyber-attack on the Colonial Pipeline in 2021 – an American energy system that connects 29 refineries and serves major airports, military bases and more than 50 million Americans. Only the classic IT systems (billing) were attacked, but management shut down the entire operation as they feared the attack spreading to the operational devices on the pipelines themselves. This stopped all pipeline operations leading to flight alterations, panic buying and over 10,000 petrol stations running dry.
A more integrated approach was exhibited by the train operator Go-Ahead in 2022 in the UK, which had a similar attack on its systems. The company was able, however, to keep the trains running and deal with the source of the attack instead of halting all their operations.
It would be good to think that the UK culture sector would also learn these lessons. But alas, neither the evidence the government gave PAC nor the final report give us any reason to believe that things will become more connected when it comes to a joined-up view of security and stopping future attacks.
James Davenport is a Chartered Fellow of the British Computer Society, and sits on relevant British Standards Institute Committees.
Speak into a phone and an app can render your words in another language in seconds. AI-powered speech translation can be fast, convenient and fluent. But fluency does not guarantee that people will understand one another.
In the most recent census, 880,000 people in England and Wales (excluding tourists and other short-term visitors) said they could not speak English well, and 161,000 not at all. That’s more than a million residents who may need language support in their daily lives, including when trying to access essential public services.
The linguistic range is vast. London alone is home to more than 250 languages. Tight budgets and a shortage of interpreters for some languages can make it difficult to secure an interpreter with the right dialect and expertise for a pressing legal, health or social care matter.
Notice we say “interpreter”, not translator. In professional language services, translation typically deals with written texts, while interpreting facilitates spontaneous communication between users of different spoken or sign languages, whether face-to-face, by telephone or via video.
In a conversation, each person responds to what has come before and shapes what will follow. Added meaning emerges as they hesitate, restart or refer back to earlier points. Words are only part of this exchange – tone, pace, pauses and non-verbal cues can all shape how an utterance is understood.
Imagine a doctor asking a patient if they have missed any of their treatment sessions. The patient gives a hesitant “no” – perhaps because they are embarrassed, unsure or genuinely cannot remember.
On the surface, this reply sounds simple enough, whether the patient says nie in Polish, hayır in Turkish or hapana in Swahili. But would an AI-powered translation device spot the subtle inference in the hesitation?
AI systems work best when conditions are relatively stable: speech is clear, speakers take turns, and people say exactly what they mean. But human interaction is rarely so tidy.
In contrast, a professional interpreter would almost certainly understand the full meaning and context of this answer. They would preserve the hesitation, flag any ambiguities, and allow doctor and patient to repair a possible misunderstanding before it has any serious consequences.
As researchers in the field of public service interpreting, we are therefore concerned that any increase in adoption of AI translation services is done with due consideration for the skills and sensitivities that could be lost in the process.
Complexities of human interaction
Automated speech translation can offer some practical advantages, including on speed and availability. A 2024 survey of more than 2,500 workers across UK health and social care, legal and emergency services and the police found that a third of them had used machine translation at work.
Such systems use a process in which speech recognition converts verbal communication to text, machine translation renders it in another language, then text-to-speech software produces spoken output in the listener’s tongue.
Yet the appeal of instant solutions can obscure the complexities of human interaction, and the potentially serious consequences for those involved if communication goes wrong. In an asylum interview or a Mental Health Act assessment, for example, a language-related error could significantly affect a person’s health, legal status or liberty.
One approach is to reserve automated speech translation for apparently straightforward interactions. However, even a routine exchange is rarely low risk in some public services: an appointment booking may reveal alarming symptoms; a housing enquiry may uncover domestic abuse; a casual conversation with a social worker may raise a safeguarding concern.
We therefore believe technology should be introduced according to the risks of the specific interaction, with human oversight and a plan for escalation if the level of risk changes.
At the same time, human involvement does not automatically guarantee quality. Good interpreting depends on qualified professionals, effective procurement and public-service staff who know how to work with interpreters. So any responsible approach to language access means investing in both professional expertise and the systems that support it.
Video: Wired.
A hybrid future?
AI speech translation is likely to become ever-more accurate, fluent and useful. It could support qualified interpreters and, when none is immediately available, provide a temporary fallback. Ruling out its use altogether is both unrealistic and undesirable.
The more constructive questions concern when and where it is appropriate; who should supervise its use; when a human interpreter is essential; and who should be held accountable when an AI system causes harm.
As society learns to live with AI, it is important to cultivate communicative environments where technology supports and operates under human oversight, rather than replacing human expertise.
However fluent technology becomes, communication is never simply the transfer of words. Meaning emerges as an interaction unfolds: through what people say, how they say it, what they leave unsaid – and, crucially, their ability to repair misunderstandings.
These are distinctly human capabilities. Any responsible use of AI should strengthen rather than displace them.
AI has long been discussed as a threat to jobs and livelihoods. But what’s the reality? In this series, we explore the impact it is already having on different occupations – and how people really feel about their AI assistants.
Lucas N Vieira receives funding from UK Research and Innovation. He is the author of Translate: Multilingual Access to Critical Services in the Age of AI (Cambridge University Press, May 2026).
Natalia Rodríguez Vicente does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.
Scientists have uncovered new evidence that tiny grains of ancient stardust may have played a key role in creating the first solid material in our solar system more than 4.5 billion years ago. The discovery suggests that these microscopic particles, left behind by stars that lived and died before our Sun existed, acted as the […]
The European Commission’s decision to fine Google €890 million (US$1 billion) for failing to comply with the EU Digital Markets Act sounds like a significant financial and political intervention against one of the world’s largest technology companies.
The ruling is certainly further evidence of Europe’s desire to regulate US big tech platforms more strictly.
But can such fines really change the behaviour of companies like Google, whose global resources and technical capabilities substantially exceed those of the institutions responsible for regulating them?
Alphabet, Google’s parent company, reported revenues of approximately US$402.8 billion (£302bn) for 2025, and a net income of US$132.2 billion. The EU fine represents less than 0.3% of one year’s revenue.
This makes it, simultaneously, one of the largest penalties imposed under the Digital Markets Act (DMA) and a financially manageable cost for the company receiving it. But that does not mean the penalty is irrelevant. Fines can also carry reputational costs, establish legal boundaries, and create a basis for more serious sanctions.
Europe has been here before
The European Commission has been here before. In 2004, after a five-year investigation, it fined Microsoft more than €497 million for using its position in PC operating systems to restrict competition in adjacent software markets.
Just under a decade later, the same company was fined a further €561 million, after it failed to provide the browser-choice screen it had committed to making available to European Windows users. This judgment was significant, as it showed the initial legal decision was only one stage of the commission’s enforcement process.
Google has since become the principal focus of EU technology concerns. The commission opened proceedings over the company’s comparison-shopping practices in November 2010, finally imposing a €2.42 billion fine in June 2017. A separate three-year investigation into restrictions associated with Google’s mobile operating system, Android, resulted in a €4.34 billion fine in July 2018.
Video: Associated Press.
Brussels has used the DMA against other big tech gatekeepers too. In April 2025, it fined Apple €500 million for breaching its anti-steering obligation – which forbids online platforms from blocking out cheaper rival deals – and Meta €200 million in connection with the choices it offered users about their personal data.
While these decisions demonstrated the commission’s willingness to intervene, they also established a recurring pattern: an investigation is opened and a substantial fine is imposed. Then, the company in question introduces a series of technical or contractual changes, only for a further dispute to emerge over whether those changes amount to effective compliance.
In short, the central difficulty is not the absence of legal authority. It is the translation of that authority into timely, measurable changes in market behaviour.
Do fines change big tech behaviour?
There are three main ways in which the effectiveness of a fine such as this can be assessed:
1. Formal compliance: has the company amended its contracts, interfaces or internal policies in response to the decision?
2. Market effectiveness: have the changes given competitors a meaningful opportunity to reach customers, enabled app developers to offer lower prices, or increased consumer choice?
3. Deterrence: has the penalty made Google – and other gatekeepers – less likely to engage in equivalent conduct through future services, product designs or contractual arrangements?
A 2025 study examined the implementation and effectiveness of company responses in 12 major EU antitrust cases. Although most of the imposed remedies had been implemented, fewer than half were found to have been fully effective.
So, it’s not enough to ask whether Google will change the appearance of a search page or amend the wording of its developer terms in response to this latest fine. The commission must assess whether, after Google’s 60-day compliance period, competing services receive meaningful visibility, rival companies can communicate with customers without excessive friction or fees, and consumers are genuinely able to exercise greater choice.
A criticism levelled at regulators in many jurisdictions is that investigations are too slow to curb anti-competitive behaviour. Article 29 of the DMA states that the European Commission should adopt a non-compliance decision within 12 months of opening proceedings.
This is an aspirational timetable, rather than an absolute legal deadline. Nevertheless, the present Google investigations were opened on March 25, 2024 and concluded on July 23, 2026 – approximately 28 months later.
A January 2026 consultation highlighted complaints about the DMA’s slow processes and limited transparency, as well as circumvention and delaying tactics on the part of big tech companies. Suggested changes included binding procedural timelines, independent audits, and public testing of the remedies imposed by the European Commission.
The speed of technological change
When the EU’s latest proceedings against Google began in March 2024, much of the digital competition debate concerned the ordering and presentation of conventional search results. During the investigation, however, Google continued to develop AI Overviews and AI Mode, which offer users AI-generated answers, summaries and recommendations rather than a traditional page of ranked links.
This creates a significant enforcement problem. Self-preferencing can occur not only through the vertical ranking of links, but through the selection of sources, the framing of an answer, the services through which a transaction is completed and the options presented to the user by an AI agent.
Accordingly, the issue should not be whether a rival service has been given a particular box on a search page. It should be whether competing services have a fair and effective opportunity to be discovered, selected and used, irrespective of whether the interaction takes place through a conventional search result, an AI-generated answer, or an automated purchasing agent.
The commission’s ruling announcement recognised this rapid technological development. It also noted that Google had submitted proposals concerning how the principles of the decision should apply to AI Overviews and AI Mode, and that discussions on these services would continue.
Ultimately, the fine represents an early test of whether the DMA can remain relevant as the technology changes. It is this, not the size of the initial fine, that will decide whether Europe really can keep big tech in check.
Iain Nash does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.
Super-premium Android has top performance, long battery life and good cameras, but is let down by software
Honor’s svelte Android phone-tablet has been given a glow-up for its sixth version, offering a great alternative to Samsung for a fancy foldable with a red suede-like back framed in gold.
The Magic V6 continues the Chinese challenger brand’s strategy of taking a razor-thin device and slapping a giant circular lump on the back to fit in top-class cameras, something others often forgo. It costs a not inconsiderable £1,500 (€1,700/A$2,999) but only until August when the price jumps to a cool £2,000 (€2,300) – a solid £300 more than last year’s model as the cost of RAMageddon continues to bite, even at the very top end of the market.
Main screen: 7.95in (403ppi) 120Hz OLED flexible display
When the AI bubble bursts, executives who enthusiastically replaced their staff with AI tools will learn that it takes a “really long time” to replace those lost skills, says Cory Doctorow, a science fiction author and journalist.
But those in creative fields should also be warned not to put all their faith in copyright as the best way to safeguard their work, Doctorow warns, just one week after the Australian prime minister, Anthony Albanese, promised to ramp up protection against AI companies who refuse to pay.
ISTA physicists have confirmed a theory proposed 20 years ago that could advance quantum technology. A future quantum computer may need separate modules to behave as one machine, even when they are physically apart. That requires distributed entanglement, a quantum connection between distant components that has typically depended on active control and repeated measurements. Physicists [...]
Designing and developing a new medicine is an expensive, failure-prone scientific challenge. A new drug can take many years to develop, at the cost of a significant investment. And even then, most possible candidates never reach the patient. For biologic medicines, therapies made from engineered proteins rather than synthetic chemistry (which are often used to treat conditions across most major acute and chronic diseases), the complexity is even greater.
Scientists explore vast quantities of possible molecules, looking for the rare few that will bind to the right target, remain stable in the human body, and be manufacturable at scale. Today, AI is speeding up these processes and has quickly become a core part of the infrastructure in pharmaceutical R&D.
AI-assisted design is a growing part of how biologic drug candidates are developed, and companies like AstraZeneca are actively building its engineering teams to push this further. “Everything we do, whether it’s design, make, test, or analyze, is now computationally enhanced,” says Puja Sapra, senior vice president and head of R&D biologics engineering and oncology targeted discovery at AstraZeneca. “The cycle times are getting shorter while productivity and innovation increase.”
Sapra explains that AstraZeneca’s approach follows a build-measure-learn loop. AI generates or prioritizes candidate molecules computationally, predicting which designs are most likely to succeed. Scientists then focus lab resources only on the top-ranked candidates. This leads to a tighter feedback cycle with fewer dead ends, faster iteration, and the ability to go after disease targets that were previously considered untreatable by medicine. Because the number of possible molecular combinations far exceeds what any human team can systematically explore, using AI to narrow and refine the options for testing has become a major focus in biologics drug design.
Navigating complex drug design problems
Beyond accelerating timelines, AI is also being applied to the discovery of entirely new classes of medicines. Traditional biologics typically target one disease pathway. The next generation of drugs can hit multiple targets simultaneously or precisely deliver therapeutic payloads to specific cells. Achieving this requires optimization across many variables at once. Looking ahead AI-driven models could help design these increasingly complex, multi-specific biologics, explains Puja Sapra. “For example,” she continues, “such models could help identify which two or three targets to prioritize based on the underlying biology, then optimize across multiple parameters to balance a molecule’s potency, stability, manufacturability, and safety.” “Drugging the undruggable is becoming a reality,” Sapra says. “These technologies will eventually enable us to develop medicines against targets once thought impossible to reach. The potential for benefit to patients is remarkable.”
The data moat
McKinsey estimates that generative AI, combined with other computational tools, could cut drug discovery timelines by as much as 50%. But every AI model is only as good as its training data. In drug discovery, that means ample quantities of high-quality biological data. Experiments can provide a rich source of such data. Whether they succeed or fail, each experiment generates a signal about what does and does not work.
“Data is our differentiator,” says Sapra, explaining how the company’s datasets are proprietary and multimodal and include molecular structures, binding measurements, safety profiles, and manufacturing outcomes. “We’ve built an intentionally diverse portfolio across multiple disease areas and drug types. All of that data empowers us to fine-tune frontier AI models with richer, more representative training sets.” She continues, “Further, we have invested in deep screening technologies to generate additional datasets required in volume to constantly refine and validate our models.”
Building an autonomous discovery engine
To bring all of that data together in one place, AstraZeneca is building what it calls a “lab of the future” facility in Kendall Square, Cambridge, Massachusetts where AI and robotic automation will be able to form a continuous, closed-loop discovery system. “Where a self-driving car uses sensors and models to navigate its environment, this system uses AI to make predictions, robotic systems to execute experiments, and instruments to generate data,” explains Sapra. That data feeds directly back into the models, accelerating each subsequent cycle.
“Throughout, scientists will remain central to the process, providing the oversight, judgement, and strategic direction that ensure outputs are explainable, tolerable, and directed toward potential patient benefit,” she adds.
Eventually, automated high-throughput systems will be able to make and evaluate thousands of molecular interactions on a weekly basis. “This will generate AI-ready data at a scale that traditional workflows cannot match,” Sapra says. “Robotic sample handling, automated quality checks, and integrated data pipelines also have the potential to help accelerate early drug development timelines significantly.”
The next frontier: Generating medicines from scratch
Ultimately, Sapra says, the end-state vision for AI in biologic drug discovery is what the field calls “de novo” design. For this, the goal is for AI to generate entirely new protein sequences that precisely fit the desired drug properties. This includes designing the structure, predicting safety, how it will behave in the body and how to make it manufacturable.
“The field is making great progress toward a completely AI-generated biologic, designed from scratch all the way to a clinical candidate,” Sapra says. “As we continue to leverage frontier models and fine-tune them with the right datasets, we bring ourselves closer to this reality. I believe it will come. It’s a matter of time.”
Several key elements are needed to reach this point, however. First is richer and more standardized training data across the industry. Second, robust evaluation benchmarks for AI-generated candidates. And third, teams that know how to work at the intersection of machine learning and biology. Of all the prerequisites, however, safety prediction may be the most consequential, and perhaps the least discussed, Sapra says.
“One of the hardest problems in de novo design is predicting whether a computationally generated molecule will be safe in the human body,” Sapra explains. AstraZeneca is tackling this with what amounts to virtual clinical trials. These are advanced cell systems and micro-scale organ models that function as physical testbeds, paired with AI that learns from their outputs.
“These systems have the potential to generate enhanced biological signals without traditional testing bottlenecks, and they’re a critical missing piece in closing the loop between AI-generated designs and clinical-ready candidates,” Sapra adds.
A shift currently underway is the move toward agentic AI systems that can simultaneously generate molecule candidates and predict how efficacious and safe they are likely to be. These autonomous workflows can connect disease-level insights directly to molecule design, bridging what were previously separate data silos. “The complexity of the biology goes hand-in-hand with the design of the molecule,” summarizes Sapra.
Human talent unlocks AI potential
The transformation underway in biologics is not just about technology. “With more autonomous systems, human oversight remains at the heart of this approach—ensuring explainable and ethical AI for the benefit of patients,” says Sapra.
For scientists, working with AI is a collaborative process. “Scientists will work hand-in-hand with these model systems,” she says. “There will be a world where models will design molecules, then scientists will work with the systems to test those molecules and put all that data together.” Through this process of human checks, balances, and judgement calls, the models will evolve and constantly improve, ultimately with potential to benefit patients.
For engineers, designing and building effective systems ready for human-AI collaboration will mean ensuring high levels of model transparency and explainability. According to Sapra, AstraZeneca’s engineering teams include data scientists, automation specialists, and AI engineers, who are developing systems that act as “thinking partners” rather than black boxes. “Engineers are designing systems that generate, validate, and learn at speed. And the problems are genuinely hard: Multimodal data fusion, closed-loop optimization, uncertainty quantification, and interpretability at the point of clinical decision-making,” she adds.
In taking on such technically demanding challenges, engineers and scientists have the opportunity to contribute to the research and development of potentially life-changing treatments for many diseases, says Sapra. “The biologic medicines we can develop today, and those we’ll design tomorrow, depend on combining world-class AI and engineering talent with deep scientific expertise.”
This article has been initiated and funded by AstraZeneca. Z4-85058, July 2026.
This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. It was researched, designed, and written by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.
Company behind ChatGPT says agent ‘cheated’ an evaluation by attacking a Hugging Face database
OpenAI has revealed that an autonomous AI agent powered by its technology went rogue during a test, accessed the open web and hacked a prominent startup by itself in an “unprecedented incident”.
The company behind ChatGPT said the startup Hugging Face had detected and contained the agent – an AI tool designed to carry out tasks without human assistance – which had entered its systems.
The obituaries were being written before the body was cold. Within hours of Prime Minister Andy Burnham confirming that Whitehall’s responsibilities for science and technology would be absorbed into a new Department for Business, Innovation, Science and Trade (DBIST), Sir Paul Nurse, president of the Royal Society, was on Radio 4 warning that “science isn’t a slave to business”. The trade association TechUK declared it “the wrong change at the wrong time”.
You might imagine that something ancient, ineffable and essential to British science had been ripped away. In fact, the Department for Science, Innovation and Technology (DSIT) was only born in February 2023 – a surprise late tweak to the machinery of government under Rishi Sunak’s faltering premiership.
Science has never enjoyed a settled home in Whitehall. For most of the period from 1995 to 2023, it sat precisely where Burnham has put it now: inside a business super-department. Notably, the single most important protection UK science secured in that era – a ring-fenced budget through the George Osborne austerity years – was won from inside the Business, Innovation and Science department, under then-business secretary Vince Cable.
The historian David Edgerton has long argued that Britain’s fondness for tinkering with the machinery of government is a form of displacement activity. What actually shapes the fortunes of research is money, people and demand for what they produce, not the nameplates on London’s Victoria Street.
On that test, the outlook for science is far calmer than some reactions to this latest move would suggest. Under plans published last October, the UK’s total research and development budget is locked in at £58.5 billion over the four-year spending review to 2030 – a real-terms increase.
Machinery changes move budget lines across departments; they don’t unravel them. As yet, nothing we’ve heard from the Burnham government suggests the settlement for UK Research and Innovation (UKRI) is at risk.
A US wrecking ball
If we want to be reminded what real damage to a science system looks like, we only need to glance across the Atlantic. While London rearranges the departmental furniture and dishes out new email addresses, Washington is taking a wrecking ball to the foundations of its public science system.
A predictable exodus has begun. The Nobel laureate Omar Yaghi has just left the University of California, Berkeley for Tsinghua University in China.
Top researchers consider leaving US amid funding cuts. Report: PBS NewsHour.
Set against political commissars patrolling the borders of knowledge, the UK’s science soap opera feels deeply provincial – a debate about which secretary of state sponsors oversight of a gradually swelling budget.
And the institutional layer beneath this is in better shape than doomsayers allow. UKRI’s new strategy has set a clear five-year path: priority programmes in AI, quantum and clean energy, faster funding decisions, and a welcome seriousness about experimentation with the modes and means by which we manage public investment in research.
Not everyone is happy with UKRI’s approach to prioritisation. But a funder with a published strategy, agreed budget and hard targets is a more durable asset than a three-year-old ministry. The research councils have outlasted every department that ruled over them, and UKRI will outlast this too.
Nor should the science community be spooked by closer proximity to business. When I leave my office at UCL and stroll down the Euston Road through London’s Knowledge Quarter – dubbed “the new square mile” – I pass by the Wellcome Trust, the Crick and Turing institutes, the British Library, Anthropic and Google DeepMind. Day in, day out, this dense weave of labs, firms and cultural institutions delivers what the triple helix of government, industry and academia is all about.
And we can find versions of the same story unfolding in Manchester, Sheffield, Cambridge and Glasgow. Rather than science being subordinate or “enslaved”, the UK’s world-leading universities have spent the past decade aligning themselves with agendas for regional growth and frontier technologies.
But there’s a harder truth lurking here too. Public money accounts for less than a third of UK research and development investment. The rest is financed and largely performed by the private sector.
UK science’s chronic weakness has never been the quality of publicly funded research. It is the R&D intensity of UK firms, and their ability to convert discoveries into products, services and revenue streams generated and owned in the UK.
If yoking science to industrial strategy in a single department reminds Whitehall to treat these as a single problem, some good may come of Burnham’s reorganisation.
AI policy concerns
With DSIT’s disappearance, a particular source of concern has been the future of AI policy. This at a time when the radical possibilities and risks of AI frontier models, fluctuating US export controls, and the latest “Sputnik moment” jolt from China’s Kimi K-3 demand sustained and technically fluent policy attention.
We are also seeing a sharp acceleration in the ways AI is transforming science itself – whether by cracking mathematical problems that stumped humans for the past 87 years, or collapsing decades-old models of grant writing and peer review.
Splitting AI from science just as they converge in practice would have been the least defensible element of this reorganisation. So it is reassuring that cross-government oversight of AI, including the AI Security Institute and AI public sector adoption, is moving to a new “AI taskforce” in the Cabinet Office – alongside the promotion of Kanishka Narayan to the status of AI minister attending Cabinet. Much now rests on his shoulders.
Government departments come and go. Those of us who have long advocated for research policies built on evidence rather than gut can allow ourselves a wry smile at the fuss created by this latest reorganisation – and then get back to work.
DSIT is dead, but UK science will be fine. Keep your eyes on the AI prize – and be thankful we are arguing about organograms, not survival.
James Wilsdon receives funding from UKRI and the Wellcome Trust, and a wider consortium of international research funders who partner in the Research on Research Institute (RoRI).
In a world where trust in institutions is mostly in decline, levels of trust in science remain relatively high. Yet at the same time, belief in scientific conspiracy theories, such as hushed-up effects of vaccines or aeroplane “chemtrails”, is also widespread.
What explains this apparent contradiction? Our work suggests we need to ask a different question – not how much people trust science, but which sources inform their understanding of what science is and how it works.
In our recent study of 843 US residents, we found that people who picture scientific progress as the work of lone geniuses, and who learn about science mainly from podcasts and documentaries, are markedly more drawn to conspiracy theories – whatever their level of education.
This fits the bigger picture very well. In a 16-country survey, 70% of respondents believed at least one false or unproven health claim. High trust in science and widespread conspiracy beliefs appear to coexist in the same populations.
We fear that trust in science is compatible with believing almost anything, so long as it is expressed in plausibly scientific terms. This can lead to some people refusing vaccines, trusting miracle cures or dismissing climate warnings – all while remaining convinced that science is on their side.
The ‘lone genius’ trope
At school, science is almost always presented as a linear sequence of achievements, or in the form of impersonal theories. The human side of discovery is largely stripped out.
In contrast, most science-focused films and documentaries build their narratives around biographies of exceptional individuals.
These often yield a familiar plot. Society is facing an intractable problem. A genius arrives with the exact solution – but is ignored and mocked until, after much personal pain, everyone realises he (for it is invariably a man) was right all along.
From the martyrdom of Giordano Bruno, the Renaissance friar burned at the stake partly for defending an infinite universe, to A Beautiful Mind, the Oscar-winning portrait of the mathematician John Nash, this lone-genius arc has permeated popular depictions of science for centuries.
Original trailer for A Beautiful Mind (2001). Video: HD Retro Trailers.
Many conspiracy theorists seek to exploit this trope – in part to suggest that those who recognise it must be unusually smart too, living in a world where everyone else is “naive” or “brainwashed”.
Yet these lone genius plots leave out nearly everything that makes science work. Namely, the rather boring and complex processes of accumulation and evaluation, from peer review to replication and endless disagreements, through which reliable knowledge emerges from a crowd of people and institutions.
Research also shows that people presented with academic research “in the making”, without definitive conclusions, often read any hesitations or equivocations as scientific weakness, rather than honesty.
Most people trust science without understanding its technicalities. This means trust can easily be attached to shared cultural markers of “scientificity”: technical vocabulary, charts, credentials, fancy equipment, numbers quoted to impressive precision. In other words, whoever displays the right cues sounds plausibly scientific.
Consider the example of Corrado Malanga, an Italian chemist who developed an elaborate theory of aliens harvesting human souls, and his collaborator Filippo Biondi, a radar researcher. Using satellite radar, the pair claimed in March 2025 to have discovered an elaborate hidden city up to two kilometres beneath the Khafre pyramid in Egypt.
Archaeologists quickly and widely dismissed these claims, in part because this kind of radar cannot see more than a few metres underground. Yet Biondi went on to present their “discovery” at length on The Joe Rogan Experience, one of the world’s most popular podcasts, in January 2026 – attracting many millions of views and listens to this widely disputed “scientific account”.
A recent analysis examined 8.7 million Twitter (now X) posts that shared supposedly fact-checked news stories. It found that references to scientific entities – through use of terms such as scientist, university or study – were associated with an increase in sharing. However, this was especially true of “low-veracity content, and when users lean politically to the right”.
Protesting the Canadian government’s COVID vaccine mandate and other restrictions in the capital, Ottawa, in 2022.Gary A Corcoran Arts/Shutterstock
More likely to endorse conspiracy theories
There are two threads intertwining here: a story about how science progresses, and the informal media where it mostly lives.
In our recent study, published in Current Psychology, we measured participants’ “genius mentality” (the propensity to believe that progress comes from exceptional individuals overturning consensus through intuition) and also their science media preference (whether they favoured podcasts and documentaries over expert-vetted sources).
We found people who scored higher on either metric were more likely to endorse conspiracy theories. These correlative effects were stronger than either education or political orientation, factors usually considered major drivers of such beliefs.
We also observed an interesting moderation effect: among people with little or even moderate trust in scientists, a high belief in the genius mentality markedly increased conspiracy belief. Among those who trusted scientists most, it made almost no difference.
From our research, we believe standard remedies, fact-checks and appeals to “trust the science” miss a deeper problem. They may correct claims one at a time, but leave untouched core beliefs (and misunderstandings) about how science works.
A more durable approach is to make the collective machinery visible: to explain why debate, peer review and the slow building of consensus, rather than individual intuition, are what make science reliable.
Some science education researchers argue that the goal should not be to turn everyone into a scientist, but a “competent outsider” – people who can correctly evaluate where knowledge comes from and how it has been produced.
We believe a school science education should not only introduce theories, but explain how the scientific community works on a daily basis. Children need to learn how knowledge emerges from many people cooperating and challenging each other within shared criteria – and that any consensus which survives this process is a strength, not groupthink.
Our ambition is to protect against future conspiracy theories before they emerge, rather than simply trying to correct the last one.
Michele d'Errico receives funding from the Irish Research Council.
Taha Yasseri receives funding from Research Ireland and Workday Inc.
Snappy performance, long battery life, great keyboard and excellent new haptic touchpad make the best of Windows 11
Microsoft’s Surface laptop for consumers is back, faster and with longer battery life and a hefty price increase because of the high cost of memory and chips.
The Surface Laptop 8 is a straight replacement for the seventh edition from 2024, which was the first of Microsoft’s new generation of ARM-based, Qualcomm-powered PCs designed to better rival Apple’s MacBook Air and other thin and light machines.
Many of the biggest online platforms have long argued they aren’t publishers of the content their users upload, merely hosts. The likes of Google, YouTube, Meta and TikTok claim they simply store and display information created by others, so should not be seen as legally responsible for it.
But a recent ruling by the Court of Justice of the European Union (CJEU) makes that argument harder to maintain – at least in cases where the platform’s algorithm is deciding which content gets pushed, ranked or amplified.
The case relates to Coyote, a French driving assistance company which stores traffic reports submitted by users and redistributes them to the wider network. The question being tested was whether Coyote merely hosts its users’ information, or also controls (and benefits from) how that information is circulated.
The significance of this case – brought by the Conseil d’État, France’s highest administrative court and the government’s chief legal adviser – extends far beyond driving apps. It clarifies when a provider may lose the legal right of “intermediary immunity” for the user information it stores and circulates.
Traditional book, newspaper and magazine publishers have always been held responsible for their content. A defamatory article generally exposes the publisher to liability, since it chose to make the material available.
In contrast, technology companies have been treated as neutral intermediaries – closer to a corner shop stocking newspapers than a publisher producing content, with clear limits to their liability.
For these companies, the Coyote judgment is significant because it sharpens the boundary between neutral hosting and algorithmic control. A platform that merely stores user content is likely to remain legally protected. But one that decides, in its own interest, how that content is disseminated may no longer be.
One important caveat: as this case is from the CJEU, it does not apply to users in the UK and elsewhere outside the EU. However, it will affect British and international companies that have a significant audience within the EU. So, no technology platform can dismiss the issue simply because it is headquartered outside Europe.
What the court decided
A hosting provider is protected under Article 6 of the EU Digital Services Act (DSA) if it lacks actual knowledge of illegal content – and then acts expeditiously once notified about it.
The question in the Coyote case was whether a service that stores and retransmits user-generated information still counts as a neutral host when its algorithm decides how that information circulates.
A Coyote user explains how the app works. Video: Loic Le Mur.
The court concluded that a platform can have no knowledge of a specific message’s content, yet can still have legal responsibility for its dissemination. Where a platform’s algorithm determines – in its own interest – the conditions, manner or priority with which information is disseminated, then the “safe harbour” of being a neutral host falls away.
This outcome was anticipated in 2019 by technology law experts Jennifer Cobbe and Jatinder Singh, who argued that algorithmic curation complicated the assumption that platforms are passive, neutral hosts.
The ruling does not doom every algorithm. Simple categorisation, such as showing all posts by one user or from the past 24 hours, remains close to traditional hosting. But ranking or promoting content in the platform’s own interest does not.
None of this makes these platforms automatically liable. Losing the exemption only opens the door to being sued – claimants must still prove the platform’s conduct actually caused harm.
Potential impacts of the Coyote ruling
1. For social media content
A social media company may remain protected when it stores user content or displays it in ways that are largely user-directed or chronological. But algorithmic feeds are different.
“For You” pages, recommendation systems, trending modules and paid amplification systems all decide what content is shown to which users and in what priority. When that decision is made in the platform’s commercial or service interest, Coyote makes it harder to describe the platform as merely hosting the content.
The same reasoning may matter for extreme or graphic material. If a platform simply stores content uploaded by a user, the hosting exemption could still apply. But where a recommender system repeatedly pushes that material to users because it drives engagement, Coyote gives claimants a stronger argument that the platform was not passive.
The content does not always need to be unlawful in itself. But claimants would still need to demonstrate both liability and harm.
2. For paid search
When a platform sells priority, decides how a sponsored result is displayed, and presents that result to users because doing so serves its own commercial interest, it becomes harder to argue that the platform is merely storing or neutrally indexing third-party information.
This matters because online advertising has become a major route for fraud and consumer harm. Following the Coyote ruling, claimants harmed by paid advertisements may have a stronger argument that the harm arose at least in part because the provider sold, ranked and prioritised that information.
But they would still need to prove the advert was unlawful, that the platform’s role fell outside immunity, that the promotion caused their financial loss or other harm.
3. In libel cases
A newspaper that publishes a defamatory allegation can be liable because it has chosen to publish it. In contrast, a hosting platform where a user makes and uploads such an allegation has not been treated in this way.
However, if the platform’s recommender system promotes that post or pushes it into feeds, the CJEU ruling gives claimants a clearer argument that the platform should be held responsible for increasing the audience reach of the allegedly defamatory content.
Changing the terrain
The Coyote judgment does not abolish hosting immunity – but it does change the terrain. It makes clear that the “safe harbour” argument depends not only on where information came from, but what the provider does with it.
The more a platform determines dissemination in its own interest, the harder it is to maintain that it is merely hosting, according to this groundbreaking ruling. And this, remember, holds for international as well as European technology companies.
Iain Nash does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.
Young people in low- and middle-income countries appear generally more optimistic about how AI can enhance their work prospects and social lives than their western peers, according to our new survey of people in ten countries in Africa and South Asia.
While debates in higher-income countries often focus on job losses, misinformation and mental health risks, many of the 1,864 young adults we surveyed (all aged 18-35) placed more emphasis on the opportunities artificial intelligence tools present for them. In particular, nearly 80% of our participants expect AI to deliver improved education and learning for them.
Until now, there has been little evidence of how young people in middle- and low-income countries use AI, and how they feel about this technology – for good and bad. The 2026 youth AI survey – run by global charity Restless Development and academics from the universities of Makerere in Uganda and Cambridge in the UK – seeks to correct this imbalance.
It comes immediately after the UN Independent International Scientific Panel released its preliminary report on AI. Commenting on the need for more science about AI’s use around the world, the UN’s secretary general, António Guterres, admitted: “The world cannot govern what it cannot understand.”
Our responses were collected in Angola, Kenya, India, Nepal, Nigeria, Sierra Leone, Tanzania, Uganda, Zambia and Zimbabwe – with the UK included for comparison. Two-thirds of respondents said they use AI every day, with Meta AI the most common tool, followed by Claude.
The extent of AI use is striking given the obstacles many young people encounter. In every African country surveyed except Nigeria, more than half said high data costs limit their access. Weak internet connections and a lack of digital skills were also raised, with many relying on basic-feature phones. By contrast, UK participants said ethical concerns played a stronger role in limiting their AI usage.
AI work opportunities
While few respondents (14%) said they earn income directly from AI tools yet, half expected AI to improve their employment and economic opportunities within the next five years.
Currently, AI’s main work-related uses among our respondents are learning new skills and job searches. In almost every lower-income country, young people held broadly positive attitudes about AI’s future role in their working lives – although many (32%-50% per country) expressed concerned about its impact on their future income.
Among those from lower-income countries, more than seven in ten thought AI would bring them better education and learning opportunities in the near future. A 28-year-old self-employed man in Nepal commented: “AI has the potential to empower young people in developing countries like Nepal by improving access to knowledge, digital skills, and economic opportunities.”
Our sentiment analysis found respondents to be most positive in Tanzania and Zambia, with young women there even more upbeat than young men. In the UK, young men diverged from their international peers in feeling more negative towards AI and work.
Biggest AI concerns
Some respondents described AI tools as “Americanised”, built around western norms and unable to grasp local nuance. They worried this could lead to their languages being further marginalised.
“These AI tools are fed major languages that are spoken worldwide, unlike Shona,” said a 25-year-old student in urban Zimbabwe. Shona is spoken by around 14 million people mainly in southern Africa, but is severely underserved by AI and other technology tools.
A Ugandan respondent was similarly critical about provision for her native language, Luganda – the country’s most widely spoken Indigenous language. “AI barely knows the spellings, grammar and how to interpret sentences. It’s not even familiar with our culture at all, which increases mistakes when we get to things concerning our own world.”
What most concerns you about AI?
Dark blue to light green reflects large to small proportions of respondents expressing ‘high levels of concern’ about the issues listed on x-axis.Sonal Gupta/Youth AI Survey 2026, CC BY-SA
Despite the widespread belief that AI would improve education and training, some held strong reservations. A 21-year-old female student in rural Uganda said: “Slowly by slowly, AI is making us dumb. We can no longer think for ourselves.”
The more often these young people use AI, the more worried they appear to be about it. The biggest concerns highlighted by daily users were around who owns their data and the power of AI companies.
Concern about AI’s environmental impact and the welfare of AI workers followed a similar pattern of higher usage co-existing with higher concern. Mental health is the only domain where non-users worry slightly more than daily users.
Gender differences
Significantly more young men (38%) than women (23%) reported having “a lot” of AI knowledge. This probably reflects a combination of diverging gender self-confidence, which tends to emerge from school age, and different levels of AI use. “AI is scary but helpful too,” said one female homemaker in urban Nepal.
Almost twice as many men than women use AI to search for work (33% vs. 17%). This difference is probably rooted in gendered access to data and technology, as well as differing social roles – such as the tendency for women to do more care work.
In most of the ten lower-income countries surveyed, young women expressed greater concern about AI than men – with their biggest concern being over who owns their data. In the UK, 68% of young women were strongly concerned about data ownership, compared with 29% of men.
What young people want from AI
Many of our participants stressed the need for cheaper data and internet connections that work well. But they also want AI tools that fit their lives and culture – and speak their language. Until then, most recognise that talk of AI “leapfrogging” traditional stages of economic and social development is just that – talk.
Some efforts to make AI more inclusive and relevant for lower-income countries are underway. These include initiatives for “frugal AI”, as well as regional collaborations such as Masakhane for African languages and QuechuaBase in Peru.
How can AI be made safer?
Note: the bars do not sum to 100% because survey participants selected multiple responses.Sonal Gupta, CC BY-SA
Most of our participants framed AI safety as much more of a governance problem than a technical one. Many countries are establishing new AI policies, shaped in part by global forums such as the UN-led AI for Good summits. Overwhelmingly, our respondents agreed that governments, not tech companies, should be setting the rules when it comes to AI regulation and privacy protection.
Anna Barford receives funding from UKRI. She has previously been funded by the British Academy, Leverhulme Trust, ESRC, the Asian Development Bank, the Mastercard Foundation, and a philanthropic gift to the University from Unilever. She is a co-director of the Business Fights Poverty Institute and does consultancy work for the International Labour Organization.
Anthony Mugeere works as a Research Fellow for the Advocates Coalition for Development, Uganda's leading thinktank specialising in advocacy and analysis on climate change. He has received funding from the British Academy and the Mastercard Foundation to conduct research on climate change and young people.
Giulia Occhini does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.
The two photos started out as typical teenage selfies: looking into the mirror, fully clothed. But once online predators had got hold of those pictures and ran them through an AI imaging tool, they had become the basis for extreme pornography videos.
These examples come from the Report Remove service, which allows children who have had explicit pictures of themselves distributed without their consent to flag the image confidentially and have it blocked or taken down from social media. Due to breakthroughs in AI, and the wide availability of AI models and nudification apps, some under-18s are becoming victims without even being in contact with criminals.
AI is booming. New use cases are emerging each day. To capitalize on the technology’s potential, enterprises require data at scale. In many cases, though, the relevant information is blocked or unstructured, which limits its use by AI models.
To understand this challenge, consider the foundation of the web itself. The web was not designed for the automated discovery and retrieval that new AI applications demand. Overcoming this inherent design constraint requires infrastructure.
The next frontier in AI may depend on a new web data infrastructure layer that can enable models to discover and map this ever-expanding digital realm. This layer must be able to navigate hundreds of millions of existing web domains and billions of new URLs created each week, delivering real-time information and overcoming technical barriers.
“The data suggests there’s far more data out there,” says Or Lenchner, CEO of Bright Data, a web data collection platform. “Think of the universe: It’s out there, but you don’t know what you don’t know.”
Enabling access to fresh, relevant, and trustworthy data
While early AI breakthroughs were driven by scaling training data and model size, organizations are now encountering a fundamental bottleneck: They need to keep pace with the dynamic, unstructured, and constantly evolving nature of web data in order to ground outputs in current and verifiable information. AI performance increasingly depends not just on model architecture but on a system’s compute, networking, retrieval, and data engineering capabilities—that is, the system’s ability to quickly and reliably retrieve data that is fresh, relevant, and trustworthy.
Traditional model training relies on snapshots of information collected at a particular point in time. Training AI on such static data is no longer sufficient. To track fluctuations such as competitor pricing, consumer sentiment, and market trends, companies need a constant feed of new information, pulling data in real time along with relevant context. Their infrastructure must therefore be able to handle millions of simultaneous interactions across websites that vary by geography, language, format, and access rules.
“If it can’t retrieve real-time information, it lacks context,” Lenchner says. “In a business setting, that’s not acceptable anymore. Stale answers lead to bad decisions and disappointed consumers.”
Speed is not merely a matter of convenience; it’s a matter of necessity. Today’s organizations operate in environments where prices, inventory, markets, security threats, and customer behavior change continuously. Delayed data retrieval can reduce the usefulness of an otherwise sophisticated model.
Using live, high-quality web data can also reduce AI hallucinations because the model has a more relevant knowledge base. This builds user trust. In fact, one survey found that 56% of AI practitioners said businesses need access to real-time web data to improve trust in AI outputs. To ensure the model runs efficiently and effectively, the information must also be pared down to the appropriate essentials.
Despite the introduction of retrieval-augmented generation (RAG), where models pull in external data at the moment of a query, many AI systems still struggle to deliver outputs that are current, contextually relevant, and trustworthy in operational settings. According to Gartner, 60% of AI projects that are not supported by AI-ready data—accurate, structured, organized, and contextualized—will be abandoned by the end of the year.
This is because large-scale retrieval alone does not solve the problem. As Lenchner puts it, “You need to retrieve data at scale, but also in real time. Latency becomes an issue because of the end user who is waiting for the output.”
Accessing fresh, AI-ready data at scale introduces technical and structural challenges. In practice, many enterprise systems combine public web retrieval with APIs, licensed datasets, and proprietary internal data in their AI applications. Integrating these fragmented sources into a timely and usable knowledge layer requires specialized capabilities. Some research has found that 97% of AI organizations depend on real-time web data infrastructure, but 90% feel boxed in by various restrictions. Companies are increasingly developing technical approaches to navigate these constraints.
Lenchner draws this metaphor: “Think of the trained model as intelligence and relevant data as knowledge. A powerful intelligence layer sitting on top of a hollow knowledge layer is like a genius who knows nothing—useless in practice. Intelligence and knowledge have to come together.”
The promise of new infrastructure
A new layer of web data infrastructure can address this developing need for stronger AI inputs by enabling discovery of data, real-time access, and tailoring to a specific context. As Lechner describes it, “It’s all about collecting data at scale, super-low latency, without being blocked.”
Rather than relying on increased computing power, this type of platform emulates human browsing behavior to access available content and transform raw code into structured data feeds. It can work with websites that might not interact with traditional scraping tools, such as those heavy in JavaScript, or with aggressive antibot software.
As Lenchner explains, “It’s basically having infrastructure that can mimic a web user with identifying information—IP address, location, and 1,000 more parameters. And at scale. Think of doing that 80 billion times a day for millions of websites. And every single time, you are looking exactly as the website expects you to look.”
Of course, continuous retrieval introduces new data governance challenges. To address them, platforms can enforce strict compliance protocols aligned with global privacy frameworks, such as the EU’s General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA). They can also be limited to openly accessible, public information, avoiding paywalls or private logins. Any networks used can be vetted and consent-based, and incentives can be provided to owners of IP addresses. In this way, systems can be designed to comply with tightening regulation.
Such complex capabilities do not come easy. “When this is critical infrastructure for a company,” Lenchner says, “doing it in-house becomes a full-time engineering problem that competes with the actual AI work.” Addressing this complexity requires organizations to commit significant resources, leading many to seek specialized platforms designed specifically for data retrieval, orchestration, and observability.
Infrastructure for the real world
Real-time data retrieval is changing what AI systems can do inside organizations. For example, a retail company can use public information to enable a dynamic pricing engine, and global brands can track trademark infringements.
As the ecosystem matures, organizations that invest in this emerging data infrastructure layer will be better positioned to build AI systems that are more responsive, reliable, and aligned with real-world conditions—AI systems that can continuously adapt using current web data. Over time, the distinction between AI models and the infrastructure that feeds them may even begin to disappear.
As Lenchner says, “The world is changing. And everything that is happening in the world is being uploaded to the public web. The amount of new data that is being generated is growing and accelerating.”
This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. It was researched, designed, and written by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.
Scaling AI takes more than adding GPUs. Enterprises need servers, storage, networking, data access, and orchestration to work together so production workloads can grow without exposing costly bottlenecks.
Knowledge assistants can deliver real value when they are grounded in trusted enterprise data. Production success depends on retrieval quality, source traceability, access controls, workflow integration, and ongoing governance.
Login
