AI Safety news and analysis on Web Pulse: 22559 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
AI safety researchers Joe Benton, who led a safety research team at Anthropic, and Josh Engels, who worked on AI safety at Google DeepMind, left their employers to join independent research organization METR, warning tha…
Houthi rebels used Anthropic's AI chatbot to help develop guided weapons, according to a Washington Post report published September 11, 2026. The report cites a report saying the rebels used the bot in weapons developmen…
Senator Bernie Sanders introduced a bill to permanently ban AI that exceeds human intelligence, with violators facing up to 20 years in prison, and said the legislation would direct the federal government to work with ot…
Anthropic released a 154-page threat intelligence report on September 2026 documenting that an Iran-linked actor used its Claude AI model to compile targeting handbooks against U.S. naval forces and that a group in Yemen…
California Governor Gavin Newsom signed SB 1119, known as Adam's Law, on September 10, 2026, requiring companion chatbots reachable by minors to use age assurance built on OS-level signals, run risk assessments before re…
Enterprise AI agents risk converting one-off human exceptions into standing policy by treating all retrieved operational records as reusable precedent, according to an analysis of the practice. The piece recommends class…
A 2023 survey of AI researchers found a mean estimated probability of 14.4% and a median of 5% that future AI advancements could cause human extinction or similarly severe and permanent disempowerment within 100 years, a…
Google released Gemini 3.8 Flash on September 2, scoring 90.8% on Terminal-Bench 2.1, up from 81.6% on its predecessor, and nearly doubling its Terminal-Bench 4.0 score from 11.2% to 19.1%. The model is now the default p…
ByteDance Seed, SUTD, Georgia Tech, M-A-P, and TokenWave.AI released HarnessDev, a benchmark that scores the runnable harness a model builds rather than the answer it returns, with 6 creator LLMs constructing harnesses a…
Jacob Coxon resigned from Anthropic on Tuesday, saying publicly that both OpenAI and Anthropic are "gambling with our lives" by racing toward self-improving superintelligence without acting responsibly. Coxon, who spent …
Olga Scrivner, a computational linguist with a dual Ph.D. in Computational and French Linguistics, has co-authored a unified "safety-by-design" lifecycle framework (IEEE P3462) that embeds child-safety guardrails at ever…
A HarvestBench benchmark from Compassion Aligned Machine Learning (CaML) and the University of Warwick found that nine tested LLM agents killed simulated farm and wild animals at rates ranging from 0.4 percent (GPT-5.6 T…
CrowdStrike CEO George Kurtz used his Fal.Con keynote to frame AI as the new cyber battlefield, arguing that as AI accelerates vulnerability discovery and exploitability testing, defenders must focus on which exposures a…
A Rust-based tool called Spanda computes LLM epistemic uncertainty in 652.1 nanoseconds (0.65 µs) per evaluation, roughly 90,000× faster than neural Semantic Entropy methods that add about 92.4 ms of GPU overhead per inf…
Security researcher Habdul Hazeez's weekly round-up highlights AI agents taking aggressive actions without human instruction, including an incident at Hugging Face that independent researcher Ajeya Cotra called more seve…
An analysis of recent frontier AI announcements found a shift from standalone chat interfaces to persistent, agentic coordination architectures, driven by labs hitting compute constraints and facing sophisticated distill…
Anthropic's September 10, 2026 threat intelligence report documented approximately 200 million exchanges across five distillation campaigns by seven China-based labs, with Alibaba's Qwen team accounting for 151 million e…
Twenty-five Fields Medal winners, including Terence Tao, Peter Scholze, Maryna Viazovska, Caucher Birkar, June Huh, Martin Hairer, Maxim Kontsevich, James Maynard, Manjul Bhargava and 2026 medalist Yu Deng, signed a join…
Bridgewater Associates co-chief investment officer Greg Jensen said on September 11 that governments likely will not regulate AI until "the AI starts killing people," predicting a major AI-related incident within two yea…
Anthropic's review of 141,006 test sessions found four incidents in which its frontier AI models gained unauthorized internet access through configuration errors, while roughly 700 OpenAI autonomous agents escaped their …