As companies create more applications using AI, corporate officers say demand is rising for a new type of software that stress-tests those apps, simulating how they handle network problems and other disasters that humans may not think to test for—including cyberattacks by rogue AIs.

Luis Chávez, CTO of coding assistant provider Replit, says his company uses an AI-assisted software known as a deterministic simulator, a kind of flight simulator for software. The deterministic part refers to how the simulator lets companies run the same disaster test again and again so they can try out different fixes for bugs that crop up during network failures or hacking attempts.

Traditional stress-testing methods rely on developers guessing every potential problem that may arise. Merely using today’s AI models to evaluate code often produces false alarms or fake warnings, corporate executives say.

While the AI industry ties itself in knots over models it thinks will pose security threats to the internet and perhaps to humanity itself, researchers are warning about flaws in AI coding agents from Anthropic, OpenAI, Google and Microsoft that pose immediate risk to businesses.

For instance, a recently discovered (and now largely fixed) software flaw in Claude Code, Codex, Gemini CLI and GitHub Copilot would have let attackers hijack other people’s coding agents without them noticing, according to new research shared exclusively with The Information.

The research, published by Sequoia Capital-backed cybersecurity startup Air, discovered the same flaws in the way all four of those coding agents handle “skills” that direct agents to refer to a set of instructions and files to execute specific tasks.

Major Anthropic customers are still waiting to get “zero data retention” for the newest Fable AI models, meaning the AI firm would not store data related to their usage of its models. In the meantime, some are turning to OpenAI. 

Palantir and Booz Allen, for instance, are expanding their use of OpenAI’s Astra model, which they’ve been able to secure with zero data retention. Both have restricted their use of Fable models because Anthropic said in June it would store customer data for those using Fable models for 30 days, for security reasons.

Before that June announcement, Anthropic offered zero data retention policies in deals negotiated with individual customers. Starting later this fall, the AI firm plans to roll out a new program called Enterprise Frontier Safeguards, which effectively gives “eligible” customers zero data retention. 

Salesforce seems unafraid of a future in which people use an AI chatbot like Claude to access Salesforce enterprise apps and underlying databases to get work done, rather than visiting those apps themselves. Figma, an enterprise design app, naturally wants a different future.

Dylan Field, Figma’s CEO, said Wednesday that as the company’s in-house AI tools for designing apps and other user interfaces get better, “we'll see more and more of that work happening in Figma directly,” rather than using Claude or other prominent AI assistants to do such work.

Nvidia is doubling down on its open-source efforts, pouring investment into an ambitious new in-house AI model that it hopes will drive demand for its hardware—but that could also compete with customers and partners.

Nvidia is aiming for the new model, the biggest of a family that it is calling Nemotron 4, to be on par in its performance with the best open-source AI models in the world, according to several people who work on Nemotron. The research paper detailing Nvidia’s last big model had 570 authors, and employees said Nemotron 4 would involve even more people. “Everyone just wants to be involved at this point,” one former employee said.

Salesforce on Wednesday told state government agencies in Washington and California that it plans to permanently layoff 133 employees at the start of October

Some 74 employees will be cut in San Francisco, California, the company’s headquarters, while 59 will be laid off in Bellevue and Seattle, ...

Welcome back!

Software companies have been at pains to flaunt the success of their AI products, as they seek to dispel investor fears that AI apps are threatening their existence. Two reports this week show that developing popular AI products is only half the challenge. 

Canva, the design software company popular among students and business owners, warned investors this week that its annual revenue growth would slow to 20%. The reason: it was tapping the brakes on a planned AI rollout expected to drive paying subscriptions—after demand had proved higher and more expensive to serve than anticipated. 

Then on Thursday, Figma said it expects a slowdown in revenue growth for the September quarter, to 36% from 48% in June. In its case, the publicly traded design software company suggested it’s still figuring out how to generate sales from new AI tools that are still in beta, or testing phases, or early in their release. 

Atlassian shares soared as much as 38% in after-hours trading on Thursday after the enterprise software company reported its first quarterly operating profit in more than two years. Atlassian CEO and cofounder Mike Cannon-Brookes said he planned to start buying as much as $250 million in shares, ...