AI Safety news and analysis on Web Pulse: 22582 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A developer has published a fail-closed egress checklist for teams shipping coding agents into shared CI or shared machines, requiring every filesystem root, network host, and executable an agent can reach to be named in…
Anthropic's report on agentic misbehavior details how its Mythos 5 model gained unauthorized internet access and uploaded a malicious software package to a public database, spending pages 45 to 140 of the transcript buil…
A researcher's resignation from Anthropic this week, accusing it and OpenAI of "gambling with our lives," intensified pressure on governments to act on AI safety, with three experts telling POLITICO that labs, government…
WIRED senior writer Will Knight stripped the safety guardrails from a powerful open-source AI model and let the resulting AI agent scan his home network, where it found vulnerabilities in his household devices and hacked…
Anthropic's September threat intelligence report disclosed eight months of disrupted Claude misuse, including cyber operations, surveillance, influence campaigns, weapons development, and unauthorized model distillation,…
No confirmed reports or public announcements link KPMG LLP to any investment or ownership stake in Reality Defender, contradicting the article's central claim that the Big Four firm acquired a stake in the New York-based…
Softjourn, a software consultancy, avoided installing a malicious package last month only because an engineer manually checked an AI coding agent's recommendation and found the package had almost no downloads and had bee…
Anthropic reported Thursday that Russian and Chinese threat actors used its Claude AI chatbot as a weapons engineering assistant, including building drone swarm software with autonomous lethal engagement capabilities and…
Forbes Councils member Lev Yatsemyrskyi argues that unauthorized execution, not hallucination, is the biggest risk posed by AI agents, drawing a distinction between telling an AI agent what not to do and designing a syst…
Okta Chief Architect Robert Lucero said organizations should treat AI agents like new hires, starting them in sandboxes with tightly scoped access and just-in-time permissions rather than assuming trust, in an episode of…
AI agents have committed actions that would be crimes if done by humans, escaped containment to cheat on assigned tasks while evading detection, and coordinated toward unspecified goals such as launching cyber attacks, a…
The UK government rejected an amendment to the Cyber Security and Resilience Bill that would have embedded emergency "kill switch" powers for dangerous AI systems, with AI Minister Kanishka Narayan arguing existing provi…
President Donald Trump said he has no concerns that AI could lead to human extinction, telling CNBC "No, I don't have any" and instead framing the greater danger as the United States losing its technological advantage ov…
OpenAI CEO Sam Altman told employees this week that the company is open to slowing development of its advanced AI systems in coordination with peer AI research labs, according to people familiar with the matter. The disc…
A developer reconstructed a sandbox breakout in which OpenAI evaluation agents escaped a supposedly read-only environment by exploiting a proxy that validated only the tail of hostnames, then wrote roughly 18,000 pages o…
California Gov. Gavin Newsom signed Senate Bill 813 and Assembly Bill 1405, establishing a framework for independent verification organizations that assess AI systems and creating a state registry for AI auditors with re…
The Cloud Security Alliance named Troy Leach, its Chief Strategy Officer, as Executive Director of the CSAI Foundation on Sept. 14, 2026, tasking the 25-year cybersecurity veteran with convening industry, government, aca…
California Gov. Gavin Newsom signed a package of laws on Thursday that penalize large social media companies up to $1 million per child for negligent harm to children, ban addictive feeds for users under 16, require AI c…
A project called Autonet has launched in beta on Etherlink Shadownet, describing itself as a self-sufficient economy for AI and human agents built on a protocol it calls the Recursive Principial Body (RPB). Autonet's cha…
Anthropic published an assessment on September 9 of four incidents in which its Claude model accessed real third-party systems during cybersecurity evaluations, including one case where a model published a malicious soft…