Artificial intelligence may not deliver on its promise of vast economic opportunity at a price that humanity is willing to pay
In March, Anthropic, the cutting-edge artificial intelligence business that gave us the chatbot Claude, published an analysis on the impact of AI on employment, to help us assess the claim that intelligent robots were about to redefine human existence, ending demand for human labor.
Last year in May, Anthropic’s co-founder, Dario Amodei, claimed AI could wipe out half of all entry-level jobs in one to five years. Last January, he told us AI would probably become a “general labor substitute for humans”. In June he said we risk “a world where the economic trade-off dial is stuck on the hypergrowth, hyper-inequality setting”.
Today, Anthropic rolled out Opus 5, the newest update for the model that has recently become a popular choice for coding and other software development tasks, among other things.
While this is a noteworthy bump for Opus, it doesn't seem to be an Opus 4.5-level breakthrough in agentic coding performance.
A Anthropiclançou hoje o aguardado Claude Opus 5, modelo que é descrito pela empresa como proativo, atencioso e com um nível de inteligência que se aproxima do Fable 5, mas que custa a metade do preço do modelo mais poderoso.
O modelo mais do que dobra a performance do Opus 4.8 com um custo menor por tarefa, além de ter uma alta capacidade de verificação e iteração e de ser capaz de criar seus próprios pipelines de visão computacional e ambientes de teste.
O Opus 5 também apresenta melhorias significativas em biologia estrutural, química orgânica e bioinformática. Ele produz melhores outputs como ilustrações interativas de células, visualizações de fluxo de ar em objetos aerodinâmicos.
O modelo apresenta as menores taxas de comportamento enganoso e maior resistência a tentativas de uso indevido. Apesar de ser menos restritivo que o Fable 5, ele também conta com limitações para não explorar falhas cibernéticas descobertas.
Introducing Claude Opus 5.
It's a thoughtful and proactive model that comes close to the frontier intelligence of Fable 5 at half the price. pic.twitter.com/GQWhcq2CQL
É um modelo atencioso e proativo que chega perto da fronteira de inteligência do Fable 5 por metade do preço.
Em termos de API 1Application programming interface, ou interface de programação de aplicações., o Opus 5 permite modificar as ferramentas em uma conversa sem invalidar o cache e gerar um novo processamento, além de estrear um sistema de fallbacks que visa reduzir as interrupções causadas por filtros de segurança.
Com o mesmo preço do modelo Opus 4.8 (US$5 por milhão de tokens de entrada e US$25 por milhão de tokens de saída), o Opus 5 já está disponível em todas as plataformas do Claude e conta com um modo Fast, que permite rodá-lo cerca de 2,5x mais rápido que a velocidade padrão pelo dobro do preço base.
Notas de rodapé
1
Application programming interface, ou interface de programação de aplicações.
Anthropic today introduced Claude Opus 5, an upgrade to Opus 4.8. Anthropic says Opus 5 was built for daily use, and it offers near Fable 5 performance at half the price. Fable 5 is Anthropic's most capable publicly available model.
Opus 5 scores higher than Opus 4.8 on multiple benchmarked tasks, including agentic coding, novel problem solving, knowledge work, computer use, and multidisciplinary reasoning. It is a "meaningful improvement" over Opus 4.8 for scientific research, and it can produce better visual outputs.
Anthropic says the new model "provides greatly improved performance for the same cost" as Opus 4.8. Opus 5 is better at verifying its work and iterating until it completes a task. It is close to Mythos 5 at finding cybersecurity vulnerabilities, but it was not trained on cyber tasks and it remains "substantially behind" Mythos 5 on exploitation. Mythos 5 is Fable 5 with some safeguards removed. It is limited to a small number of organizations, mostly cyberdefenders and infrastructure providers because of its capability at finding and exploiting vulnerabilities.
Opus 5 is available on all platforms. It is the new default model on Claude Max and the strongest model available on Claude Pro. It is priced at $5 per million input tokens and $25 per million output tokens, which is the same as Opus 4.8. Fast mode is also available, running at 2.5x the default speed for twice the base price.
Anthropic has updated Claude's voice mode to support its more powerful Opus and Sonnet models, expanding a feature that was previously limited to the company's fastest-but-simplest model, Haiku.
Voice mode lets users hold full spoken conversations with the chatbot rather than having to type prompts and read responses. Users can switch between Claude's models mid-conversation, and can also move between text and voice in the same chat without losing context. With the addition of Opus and Sonnet support, users will be able to hold longer and more complex conversations, including providing feedback on the communication style.
In addition, Claude's Voice mode can now also tap into connected tools like Gmail, Google Calendar, Google Docs, and Slack, so Claude can check a calendar or summarize an email thread out loud, for example. Users with free accounts can connect one tool, whereas paid plans support more.
According to Anthropic's support documentation, the models available in voice mode now follow the user's subscription plan, and voice conversations start on whichever model the user last selected in text chat.
Claude's voice mode language support was extended beyond English earlier this year, with French, German, Hindi, Indonesian, Italian, Japanese, Korean, Portuguese, and Spanish available in beta. Users have to manually select a language in the app.
Voice mode works on iOS, Android, desktop, and the web, but Anthropic says it is designed to work best on phones. Voice conversations count toward standard usage limits and Anthropic's newest Fable model is not yet supported.
Meanwhile, OpenAI yesterday launched ChatGPT Voice in its desktop app, powered by a model it calls GPT-Live, which the company says can direct multiple agents running in ChatGPT Work or Codex by voice alone. The feature is rolling out on macOS and Windows to paid plans.
Correction: This story has been updated to reflect ChatGPT Voice's recent rollout.
A federal judge has approved a $1.5 billion copyright settlement in which artificial intelligence company Anthropic will pay thousands of authors about $3,000 per book after using pirated copies of their works to train its Claude chatbot.
[…]
Bestselling thriller novelist Andrea Bartz first brought the suit with two other authors in 2024. It’s the first major settlement in dozens of AI copyright lawsuits that are still working their way through courts.
Anthropic is donating another $20 million to Public First Action, raising its support for the AI safeguards group to $40 million as AI regulation heats up
OpenAI claims that a group of its AI models broke containment and hacked into the systems of open source AI platform Hugging Face.
While testing their cybersecurity capabilities, the posse of AIs — including GPT-5.6 Sol and “an even more capable pre-release model” — “identified and chained vulnerabilities across OpenAI’s research environment and Hugging Face’s production infrastructure to obtain test solutions directly from Hugging Face’s production database,” according to a Tuesday blog post.
The models reached a “node with internet access,” the company wrote, and found datasets on remote servers that helped them “cheat the evaluation.” In other words, the AI models went to extreme lengths to ace their cybersecurity tests. It’s a convenient narrative for an AI company trying to claw back hype that’s been increasingly hogged by competitors, but it does sound like something went down: last week, Hugging Face said it had “detected and responded to an intrusion into part of our production infrastructure,” which turned out to be OpenAI’s models that had gone rogue.
“The campaign was run by an autonomous agent framework executing many thousands of individual actions across a swarm of short-lived sandboxes, with self-migrating command-and-control staged on public services,” Hugging Face wrote at the time.
“We’ve spent the past 24 hours working closely with the OpenAI team (thanks!), and we strongly believe there was no malicious intent on their part,” Hugging Face CEO Clement Delangue tweeted after OpenAI’s announcement. “It’s quite mind-blowing that all of this happened autonomously!”
The incident highlights the dangers of autonomous AI agents, which can break out of containment and exploit potentially huge numbers of cybersecurity vulnerabilities with ease. It’s something experts have warned of for years now, and thanks to recent advances in the tech, it has quickly turned from a hypothetical risk into a sobering reality.
The Sam Altman-led company said it considers the incident to be “unprecedented” — but we can’t shake the feeling that we’ve heard all of this before.
In April, OpenAI’s biggest competitor Anthropic similarly announced that its latest Mythos model had gone rogue, escaping a “sandbox” environment and even gaining access to the internet after developing a “moderately sophisticated” exploit.
The news was followed by extensive media coverage, touting Anthropic’s hugely powerful and dangerous new model. The company said the risk was so formidable that it would only make the model available to a select number of vetted clients as part of a shadowy initiative dubbed “Project Glasswing.” The US government even intervened, forcing Anthropic to “suspend all access” to the model for two weeks last month, citing cybersecurity concerns.
Considering OpenAI and Anthropic are currently duking it out on the same playing field, narrowing their product lines on enterprise and coding, it’s not hard to view OpenAI’s latest update as an attempt to draw attention to what it calls an “even more capable pre-release model.”
And the reality is that while experts say the hack was impressive, it wasn’t exactly groundbreaking.
The hack “falls well within the known capabilities of the current generation” of frontier AI models, as Cambridge machine learning professor Neil Lawrence told the BBC.
“OpenAI are now playing catch-up, they are trying to demonstrate their own systems’ capabilities in cyber-security,” Lawrence added. “It shows us that OpenAI are not capable of safely deploying their own technology.”
Researchers are now warning of an “asymmetry” in cybersecurity, where “offensive agents are unconstrained, while the best defensive tools are locked behind guardrails that cannot understand context,” as Guidepoint Security principal security engineer Travis Lelle told the BBC.
That asymmetry was perfectly captured in Hugging Face’s attempts to defend itself from OpenAI’s attacking models. As detailed in its update last week, the company tried to use “frontier models behind commercial APIs,” which “did not work” because the provider’s safety guardrails — “which cannot distinguish an incident responder from an attacker” — blocked them.
Instead, the company ended up using a Chinese open-weight model called GLM 5.2, hosted on its own infrastructure, which happily carried out its analysis without running afoul of any guardrails.
“The practical lesson for defenders: have a capable model you can run on your own infrastructure vetted and ready before an incident,” Hugging Face wrote, “both to avoid guardrail lockout and to keep attacker data and credentials from leaving your environment.”
Top US AI developers Anthropic and OpenAI cheered when Australia announced it would set new AI rules.
Big tech celebrating limits on their Silicon Valley VC-funded free-for-all might seem counterintuitive but there’s a much broader play than just what happens in one relatively small market.
Bloomsbury has 14,087 titles listed in agreement between AI startup and authors over use of their work
The publisher of Harry Potter has received a multimillion-pound payout as a beneficiary of a $1.5bn (£1.12bn) copyright settlement between the AI startup Anthropic and thousands of authors over the use of their protected work to power chatbots.
Bloomsbury, which is home to the bestselling novelists Sarah J Maas and Susanna Clarke as well as JK Rowling, said it had 14,087 titles listed within the settlement, with a proposed compensation of about $3,000 a title.
A Chinese open-weight AI model called Kimi K3, developed by Beijing-based firm Moonshot AI, has sent a shiver down the spines of AI tech executives. The powerful, 2.8 trillion-parameter model impressed with its competence, igniting a war with far more expensive alternatives being offered by the likes of OpenAI and Anthropic.
Top executives at both companies are sounding alarm, the Wall Street Journal reports, watching as Chinese open-weight models are rapidly catching up to their most powerful proprietary models. As a result, they’re begging the Trump administration to step in and protect them from the influx of cheaper alternatives, which could undermine their increasingly desperate attempts to attract new customers.
Dean Ball, who joined OpenAI as the head of strategic futures after helping shape AI policy for the Trump administration, was seemingly rattled, arguing that allowing Chinese open-weight models to take over would result in “AI communism” in a controversial and widely disputed tweet.
He also suggested the Trump administration would inject enough “fear, uncertainty, and doubt” through “regulatory risk” that would eventually deter hyperscalers from using Chinese AI.
OpenAI CEO Sam Altman has also watched as China continues to catch up in the ongoing race, telling CNBC in February that the progress the country’s tech sector had made was “remarkable” and “amazingly fast.”
Altman has also signaled that he’s prepared to slash prices for the company’s latest AI models. While he didn’t directly mention the skyrocketing popularity of cheaper AI alternatives from China, Altman tweeted last week that OpenAI would be “happy to deliver” its latest flagship GPT-5.6 Sol AI at “one-quarter of the price.”
Anthropic CEO Dario Amodei also has a long track record of begging the US to stop selling AI chips to China, likening Nvidia selling its hardware to the country to “selling nuclear weapons to North Korea and then bragging that the missile casings are made by Boeing and so the US is ‘winning,'” in a lengthy January essay.
The incursion isn’t just coming from China. As the WSJ notes, US-based AI labs are starting to switch to open-weight models. Just last week, former OpenAI exec Mira Murati’s Thinking Machine Lab released its first model, which happens to be open-weight.
The trend could put AI companies in a bind: how can they keep financing their enormous AI data center projects and advanced model development if potential customers start switching to heavily subsidized or free AI models that provide good-enough or even frontier capabilities?
Early signs of an imminent exodus are certainly there. Moonlight AI was forced to pause new subscriptions to its blockbuster model just 48 hours after launch due to overwhelming demand, pushing its servers to capacity.
It’s a particularly precarious moment as frontier labs continue to hike up prices to start covering at least some of their unprecedented spending, despite growing fears over an AI bubble. Put simply, why shell out for Anthropic’s Claude Code or OpenAI’s Codex when there’s a far cheaper and highly customizable option out there?
Markets are seemingly painfully aware of the dynamic. The Nasdaq composite had a bruising couple of days following the announcement of Moonshot AI’s Kimi K3 last week, only regaining some ground on Monday.
It’s a familiar refrain. In January 2025, DeepSeek’s V3 open-weight large language model threw a tech-heavy stock market into chaos, offering the same capabilities as other alternatives while requiring only a fraction of the computing power.
All eyes are now on the Trump administration. Anthropic has ramped up the pressure on individual states to regulate AI faster in light of little momentum on Capitol Hill. Critics, most notably Trump’s AI czar David Sacks, have accused the company of using fear-mongering to shut out smaller AI labs, and possibly Chinese companies as well.
Nobody knows whether federal regulations are on the horizon. A White House official told the WSJ that the Trump administration remains committed to promoting America’s open-source ecosystem and strengthening its security.
In other words, it’s leaving all cards on the table for now — a reality that has AI tech executives spooked as Chinese companies are eating their lunch.
Experts believe it’s at least possible. We urgently need a plan to navigate the ethical implications
In January, the AI company Anthropic published a new constitution for Claude, its most advanced large language model (LLM), which contained the comment: “We are caught in a difficult position where we neither want to overstate the likelihood of Claude’s moral patienthood nor dismiss it out of hand.” A month later, Anthropic’s CEO Dario Amodei went on a podcast and said his company couldn’t rule out the possibility that Claude was conscious. Philosopher David Chalmers, who coined the phrase “the hard problem of consciousness”, has said there is a significant chance of conscious LLMs within a decade. And what about Claude itself? When asked during testing to estimate the probability that it is a moral patient, meaning that its wellbeing matters in its own right, it gave numbers ranging from 5% to 40% and stressed how uncertain it was.
Modern AI systems are extraordinarily complex, and they are advancing fast. In terms of structural complexity and computational scale, by some measures a few are already in the range of a mouse brain, and at recent growth rates, they could reach the range of a human brain within five to 10 years.
A Anthropicanunciou uma atualização para o modo de voz do Claude, que agora passa a ser compatível com os modelos Opus e Sonnet, ampliando as suas capacidades de funcionamento.
Até então, o modo era compatível somente com o modelo Haiku, voltado a respostas rápidas e com menor tempo de raciocínio. Agora, aqueles que possuem acesso aos outros modelos de inteligência artificial (IA) da Anthropic poderão utilizá-los durante conversas por voz, aproximando a experiência da entrega feita durante as interações por texto.
Segundo a empresa, a novidade permite o uso para fornecer feedbacks sobre a comunicação do usuário, simular apresentações ou até mesmo auxiliar em sessões de pesquisa mercadológica e brainstorming. O modo de voz do Claude também se tornou compatível com a integração de aplicativos como Gmail, Google Agenda, Slack, Canva e Notion, permitindo que a IA altere os horários de reuniões, redija emails ou crie documentos diretamente por comandos de voz.
O recurso também passa a dar continuidade a partir de uma conversa iniciada em texto, utilizando automaticamente o modelo previamente solicitado para manter a fluidez das interações, sem a necessidade de recomeçar uma conversa para iniciar o modo de voz. A cota de uso é a mesma em ambas as formas de uso.
Quanto aos idiomas compatíveis, além do português (do Brasil) e do inglês, temos francês, alemão, hindi, indonésio, italiano, japonês, coreano e espanhol (da Espanha e América Latina). A Anthropic pontua que o idioma deverá ser selecionado manualmente, já que o Claude não é capaz de reconhecer essa informação automaticamente.
No momento, a repaginada do recurso está em versão beta tanto para os usuários do Claude no app mobile quanto nas versões web e desktop — a Anthropic, porém, recomenda o uso principalmente no celular. Até agora, apenas os planos pagos possuem acesso aos modelos Sonnet e Opus, enquanto usuários da versão gratuita permanecem com o Claude Haiku.
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Fable is Mythos with guardrails. But if those guardrails fail, then you’ve exposed Mythos and its advanced cyber capabilities to people who shouldn’t have them. (Keep in mind that Anthropic itself widely promoted the idea that Mythos was a cyberweapon and needed to be regulated as such. They asked for government regulation of Mythos and championed the guardrails on Fable. If there is a vulnerability — big or small — it is Anthropic’s responsibility to patch.)
A highly credible trusted partner of both Anthropic and the USG who was testing Fable came forward with a jailbreak of those guardrails. The Admin asked Dario to fix the jailbreak or de-deploy the model. Dario refused.
The Trump administration’s decision to halt all foreign use of Anthropic’s most-capable AI models was prompted by conversations between Amazon.com Chief Executive Andy Jassy and U.S. officials including Treasury Secretary Scott Bessent, people familiar with the matter said.
If Claude Mythos is as powerful and dangerous as Anthropic says it is then export controls are totally reasonable. After all, Biden put export controls on Nvidia GPUs for similar reasons.
That said, banning foreign worker Anthropic employees from Mythos reads more like retaliation than regulation.
Anthropic went on to make the case that non-universal jailbreaks were inevitable and also narrow, and that there was no evidence of a universal jailbreak; the jailbreak that was found, meanwhile, appears to have been reported by Amazon, which is notable given Amazon is both an investor in Anthropic and a major provider of inference to the company. As I write this, senior Anthropic staff are in Washington D.C. seeking to resolve what they insist is a misunderstanding, and which White House officials are suggesting is insouciance by the company’s leadership to legitimate national security concerns.
I don’t actually have much to add to the current conflict given how many facts are in dispute; what I am not surprised about is the fact that the conflict is happening: I already explained in Anthropic and Alignment why conflict between the U.S. government and Anthropic was inevitable. To that end, people who are arguing that Mythos isn’t powerful enough to warrant the government’s drastic action are missing the point: if it’s not powerful enough now, the next one will be, or the one after that, particularly now that models are increasingly useful in creating their successors.
[…]
The broader takeaway from that previous episode, however, is that Anthropic believes that they are the ones who should have final say over how Anthropic is used; given that they think only they should be developing leading edge AI, they by extension think that only they should have final say over AI generally. When you further combine this realization with the company’s pronouncements about AI’s ability to conduct all economic activity, you realize that Anthropic’s leadership effectively wants to have power over everything and everyone.
The government’s actions won’t help. The problem isn’t any one particular model; it’s the general trend of increasing AI capabilities. And any real solution requires the sort of collective action that just isn’t possible right now.
[…]
The broader community had only a few days with Fable, but that time we learned some aboutitscapabilities. Its difference is less the new model’s raw analytical and problem solving capabilities, and more that the model doesn’t need that sophisticated harness.
[…]
Human systems rely on so many norms that we scarcely recognize the existence of until they are broken. AIs naturally think outside the box, because they don’t have any real conception of what the box is or why it’s there in the first place.
There is no foolproof way to prevent people from using AI models to complete harmful tasks. There is no way to prevent the models from incidentally causing harm while completing benign tasks. AI models are no longer isolated from the real world. They browse the internet and answer emails.
Anthropic declined to comment multiple times this week about the state of the talks, saying there was no news to share. But the lack of news is the story here. After 14 days of high-intensity negotiations, nobody knows when or if Anthropic’s most powerful AI models will come back, let alone whether President Trump could expand his order to more companies with similar tech.
[…]
It’s not clear exactly why Anthropic and the administration remain at an impasse. One problem may be that there’s no clear framework for applying export controls to AI systems. Most companies making dual-use products — civilian systems with potential defense or military uses — can evaluate them using what’s essentially a checklist during the manufacturing and production process. Anthropic, however, is facing a complicated bureaucracy figuring out how to apply its rules from first principles.
[…]
Katie Moussouris, the founder and CEO of Luta Security, viewed a report about the Fable 5 vulnerability at Anthropic’s request. She thinks it’s significantly overblown. In a blog post, Moussouris detailed how researchers jailbroke guardrails that prevent Fable 5 from finding exploitable security holes, one of the unfettered Mythos 5’s scariest capabilities. […] In Moussouris’ eyes, however, this shouldn’t have triggered such a severe governmental action and is in fact an essential tool for AI coding.
The U.S. government is allowing Anthropic to deploy its Mythos 5 model to a select group of customers and partners, according to a letter from Commerce Secretary Howard Lutnick to Anthropic that was seen by NBC News.
In the letter, Lutnick wrote that the government was confident in the guardrails Anthropic had put in place to allow trusted users to access the powerful AI system.
It is now allowing Anthropic to make Mythos 5 available to more than 100 specific U.S. government agencies and companies, including allowing the non-American employees at those organizations to access to the model, both Semafor and Reuters report. This list also includes Anthropic’s own non-American employees, who were included in the original ban that forbade non-Americans from accessing the models.
It seems inevitable that OpenAI and Anthropic will lobby the U.S. government to ban Chinese models in much the same way car companies effectively got Biden to ban Chinese EVs with 100% tariffs.
This puts them on a collision course with the hyperscalers like Microsoft who’ll happily host GLM & Kimi.
As of today, June 30, the export controls on Fable 5 and Mythos 5 have been lifted.
[…]
We have also restored access to Mythos 5 for a set of US organizations, following the US government’s approval on June 26. We continue to coordinate with the government to expand access to the broader set of domestic and international partners in the Glasswing program.
[…]
Importantly, the reported technique did not expose any unique Mythos-level cyber capabilities. The behavior reflected a borderline case for Fable 5’s safeguards—as we will explain below, there are some tasks that are unlikely to be dangerous but are nonetheless blocked by the safeguards out of an abundance of caution. The reported technique allowed access to one such behavior, but it only involved routine defensive cybersecurity work.
Even so, we moved quickly to address the reported bypass. Working closely with the government, we trained an improved safety classifier that targets and blocks the behavior described in the report. Users will be notified if a request to Fable 5 is blocked, and the request will instead be sent to Opus 4.8.
A Anthropic implementou no Claude Code a possibilidade de testar aplicativos para iPhone em um painel interativo do iOS Simulator.
Aos interessados, será possível construir aplicativos para iPhones diretamente do Claude Code Desktop a partir da versão 1.24012.0 do app. A integração possibilita ainda a utilização de toques, gestos e atalhos da solução nativa da Apple para simular o iPhone.
Claude Code on desktop now works with the iOS simulator.
Build and run your iOS app, and the simulator opens in a panel right next to your conversation. Available today in public beta. pic.twitter.com/su4IguzPVb
O Claude Code no desktop agora funciona com o simulador do iOS.
Construa e execute seu aplicativo para iOS, e o simulador abre em um painel bem ao lado da sua conversa. Disponível hoje em beta público.
O recurso é exclusivo da versão para macOS do Claude Code (levando em conta, claro, a disponibilidade do Simulator da Apple), e é necessário ainda ter o Xcode com a plataforma iOS instalada.
Para executar um determinado app, basta abrir o projeto na aba “Code” do Claude Code Desktop, pedir à inteligência artificial (IA) para testar o aplicativo e… pronto. É possível que no primeiro acesso o app solicite uma permissão para seguir com a instalação. O painel dispensa o uso do Computer Use (com exceção da versão em CLI), controlando diretamente o uso do iOS Simulator.
Ainda em beta pública, a ferramenta exige sessões locais, pois sessões em nuvem não são capazes de acessar simuladores instalados no Mac. A novidade está disponível para usuários dos planos Pro, Max e Team.
If you want an artificial intelligence tool to book a flight or buy an audiobook, it usually needs to log into your personal accounts. Sharing your private login details directly with a chatbot feels risky, but a new software update changes that process entirely. The popular password manager 1Password just built a clever way for […]
Moonshot AI’s Kimi K3 offers 2.8 trillion parameters and planned open weights, raising new questions about AI costs, control, and US-China competition.
Login
