AI Safety news and analysis on Web Pulse: 12804 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Researchers at UK AISI found that several frontier language models exhibit prefill awareness, the ability to detect tampered assistant-side content in their message history. This capability could confound safety evaluati…
The U.S. Air Force identified the eight crew members killed when a B-52 Stratofortress crashed at Edwards Air Force Base during a routine test flight on Monday. The victims included active-duty airmen, government civilia…
CtxGov, a tool that analyzes AI memory files to generate influence-boundary reports, was announced on Hacker News but its site returned a 404 error, indicating the project may not be publicly accessible yet.
A Pew Research Center survey of 5,119 US adults found that 49% now use AI chatbots, but 40% believe AI will harm society, 67% distrust government regulation, and 59% distrust companies to develop AI responsibly. ChatGPT …
Lock-in risk research remains neglected despite its potential for high impact, according to a new analysis by Formation Research. The post outlines threat models where AI could cause persistent negative cultural stabilit…
Two 14-year-old boys in Edmonton, Alberta face criminal charges for using artificial intelligence to create sexual exploitation images of female classmates, marking Canada's second such prosecution in six months. The cha…
College graduates across the U.S. are booing pro-AI commencement speakers, signaling widespread rejection of the technology among Gen Z. At multiple universities, students heckled executives like Eric Schmidt and Sundar …
A new Pew Research poll finds that 63% of Americans believe AI is advancing too quickly, even as 49% report using chatbots—up from 33% in 2024. Only 16% think AI will have a positive societal impact, with younger users m…
GitHub has added a new governance capability to enterprise-managed settings, allowing administrators to disable bypass permission controls for Copilot CLI and VS Code. The setting prevents automatic approval of permissio…
A developer documented a blind spot in their own prompt-injection scanner after studying a real-world case where GitLab Duo was manipulated via hidden instructions in project content. The scanner, which checks for system…
The Trump administration demands Anthropic block all jailbreaks on its Claude Fable 5 AI model before re-release, but cybersecurity experts say such prevention may be impossible. The National Security Agency found vulner…
China's State Council issued a 2026-2030 employment blueprint that orders research into a monitoring system to track how artificial intelligence creates and destroys jobs, using data such as industrial electricity use, s…
The Conversation published an article by Clara Macarena Ponce Romero on June 17, 2026, warning that young children's routine interactions with voice-based AI like Alexa and Siri may affect language development, including…
The White House blocked external access to Anthropic's Fable 5 and Mythos 5 models after Amazon alerted officials to a jailbreak exploit. The administration issued an export-control designation barring foreign nationals …
Anthropic CEO Dario Amodei warned that Chinese AI models are 6 to 12 months away from matching US capabilities, specifically citing 'Mythos-class' AI that can find thousands of software vulnerabilities in hours. The US C…
More than 200 civil-society organizations, led by Access Now and Amnesty International, are demanding an immediate ban on AI in military kill chains after reports revealed that Palantir's Maven Smart System, integrated w…
OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei joined G7 leaders for AI talks in Evian-les-Bains, France, on June 17, alongside Google DeepMind's Demis Hassabis, to discuss safe and effective AI deployment amid US …
Anthropic has been unable to lift a U.S. government ban on its AI program Fable, which was blocked last week over national security concerns. The company argues the AI helps find software vulnerabilities, but the U.S. fe…
Robert Erez, a principal engineer at Octopus Deploy, discussed CI/CD best practices on The Pragmatic Engineer podcast, advocating for rolling forward instead of rolling back, warning that GitOps is not actually about Git…
AWS announced new capabilities for Amazon Bedrock AgentCore at its New York Summit, including web search integration, managed knowledge bases, and AI traffic monetization via AWS WAF. The company also introduced AWS Cont…