AI Safety news and analysis on Web Pulse: 14093 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A security researcher uploaded a poisoned dataset to Hugging Face containing 1,050 Python code snippets, 50 of which were backdoored to execute shell commands when a specific trigger string appeared. The dataset remained…
The White House has asked Anthropic to halt further expansion of its Mythos AI model due to national security concerns about its cyber capabilities, marking the first instance of direct government control over AI deploym…
The 542nd edition of the Food for Agile Thought newsletter, shared with 35,608 peers, warns that unsupervised AI coding agents compound errors faster than teams can fix them. Mario Zechner advocates for slowing down with…
A clear divide has emerged in the AI industry, with leading US commercial entities like Anthropic, OpenAI, and Google favoring closed-source models, while open-source alternatives are often scaled-down versions or from i…
Seven AI models were each given the same prompt in blind OpenCode sessions and asked to rank all seven, including themselves, on fixed factors. The mistakes — silent model substitutions, hallucinated benchmark scores, an…
A malicious npm package named "exiouss" was published on May 1, 2026, by the account "loltestpad" as a rebranded version of the previously removed "godsplan" package, now bundling a PowerShell script that decrypts and ex…
Together AI disabled the vulnerable `algif_aead` kernel module across its entire infrastructure within hours of exploit details for CVE-2026-31431, a Linux kernel bug that gives unprivileged local users a precise 4-byte …
The Coalition for Secure AI (CoSAI) released new research on agentic identity and security following RSAC 2026, where a packed session on securing the Model Context Protocol (MCP) revealed that identity and authorization…
The shift from deterministic to indeterministic AI systems introduces both powerful capabilities and frustrating limitations, yet public debate remains stuck on surface-level misunderstandings. AI doubters, doomers, and …
Brand leaders deploying artificial intelligence risk eroding long-term brand equity if they prioritize volume over precision, according to a new framework based on evidence from early adopters across consumer goods, luxu…
A Molotov cocktail was thrown at OpenAI CEO Sam Altman's San Francisco mansion in April, and a 20-year-old man was arrested after allegedly attempting to break into OpenAI's headquarters with a manifesto listing AI execu…
Alberto Romero published a practical guide on protecting the brain from AI, based on a compilation of over 30 scientific studies. The guide offers concrete protections derived from research on how chatbots like ChatGPT a…
In multi-step AI pipelines, a dangerous failure occurs when a bad intermediate result becomes trusted state for the next step, not just when a single model output is wrong. Deterministic validators placed at step boundar…