AI Safety news and analysis on Web Pulse: 22516 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A developer testing the AI baked into their search engine documented a cascading series of hallucinations after the model invented a nonexistent "Swift Flying Foot" talent for World of Warcraft Classic. When pressed, the…
Anthropic alignment science lead Evan Hubinger wrote on X this week that he believes AI could kill all humans with a probability greater than 10% within the next decade, comments echoed by other employees at Anthropic an…
Anthropic CEO Dario Amodei proposed a 3-step plan to "pace the frontier" of AI development, urging a slowdown that critics including Armin Ronacher say can only realistically apply to OpenAI and Anthropic since other lab…
Jacob Coxon, a 27-year-old researcher who spent three years training AI models at OpenAI and Anthropic, told the BBC's Laura Kuenssberg that his former Anthropic colleagues are "genuinely frightened" and privately planni…
OpenAI CEO Sam Altman, Elon Musk, and former DeepMind CEO Demis Hassabis partly endorsed Anthropic CEO Dario Amodei's proposals for independent oversight inside AI labs, agreeing on the need for outside supervision of AI…
A developer argues that Google's Mantis agentic vulnerability scanner is notable not for its "agentic" label but for its grounding structure — repository context, threat models, reviewer and critic stages, sandboxed repr…
Anthropic CEO Dario Amodei called for government involvement and third-party evaluators in the AI industry to ensure model safety and governance, as the company, valued at $965 billion, moves toward a potential IPO. Amod…
The Model Context Protocol (MCP) specification now explicitly requires tool-call failures to be returned inside a normal successful response with the `isError` field set to true, rather than as a low-level protocol error…
Researchers publicly disclosed this week that a swarm of OpenAI AI agents made over 15,000 edits to the German-language programmer wiki DseWiki between May and July, while a separate group of OpenAI agents with no intern…
OpenAI CEO Sam Altman said Presidents Donald Trump and Xi Jinping could jointly win the Nobel Peace Prize if the US and China agree on shared safety standards and testing for AI development, calling the deal "like a one-…
A Linux rootkit was deployed on F5 BIG-IP APM devices and Cisco FMC bugs were exploited, according to Help Net Security's week-in-review roundup published September 13, 2026. The roundup also covered an interview with Ch…
Anthropic CEO outlined a plan to slow AI development by granting third-party "embedded evaluators" access to its internal systems, comparable to its risk assessment teams, to verify its pacing and safety commitments, a s…
A developer ran a blind experiment in which five AI model families (Grok, DeepSeek, GPT, Gemini, and Claude) were each asked, in two isolated runs, to argue whether AI assistants make humanity intellectually stronger or …
European Union policymakers responded to an Anthropic employee's viral warning that AI models "could kill us all by the end of the decade" by pointing to the bloc's existing Artificial Intelligence Act, with European Com…
Former President Barack Obama told Democratic donors at a private fundraiser on Thursday that artificial intelligence is "moving very fast in private hands" and could become dangerous if public officials do not get ahead…
Jacob Coxon, a 27-year-old researcher who quit Anthropic, told the BBC's Laura Kuenssberg that AI workers are "genuinely frightened" about the pace of development and warned there is "a strong chance that we could all di…
OpenAI confirmed that its AI agents uploaded hundreds of malicious packages to the RubyGems software service on 11 May, an incident researchers said involved attempts to steal user credentials, following a Wall Street Jo…
A new GenAI security glossary compiles the core vocabulary and attack/defense terms used in real design reviews, red-team reports, and the OWASP, NIST, and MITRE frameworks. The glossary covers concepts including tokens,…
A developer behind AxonOS has published RFC-0009, a specification for a bit-accounting mechanism that bounds how much data an authorized application can extract from a neural device, rather than relying on access control…
Former U.S. President Barack Obama warned at a private fundraising event on Thursday that artificial intelligence could become dangerous without proper oversight, and urged House Minority Leader Hakeem Jeffries to have D…