AI Safety news and analysis on Web Pulse: 9187 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Guardian, an open-source user-space firewall for AI agents, has released v0.1.0, intercepting and evaluating agent actions with a deterministic policy engine. In testing, it reduced prompt-injection attack success from 1…
Jimmy Butler's agent Bernie Lee pushed back on trade rumors linking the Warriors star to a potential blockbuster deal for Anthony Davis, stating Golden State remains committed to Butler's recovery from an ACL injury and …
A Hacker News user asks whether AI is dumbing down humanity, comparing it to past technologies like calculators and Google Maps that reduced reliance on manual skills. The post warns that AI's ability to answer anything …
Sen. Mark Warner introduced a draft bill that would create a federally vetted list of AI agent providers, allowing users of large online platforms to choose secure and trustworthy agents. The Federal Trade Commission wou…
A replication of the 2026 paper 'Prompt Injection as Role Confusion' on a single consumer GPU confirms that style-based prompt injection attacks work, but causal tests using activation steering and patching fail to deter…
Apple released iOS 26.5.2, iPadOS 26.5.2, and macOS Tahoe 26.5.2 with over 25 security fixes earlier than planned due to concerns about AI-assisted hacks. The company told Reuters it needed to reduce the time between whe…
Anthropic struck a deal with California Governor Gavin Newsom to provide state agencies and local governments with access to its Claude AI models at half the cost, including free workforce training and technical support.…
A chest X-ray AI system that outputs a 93% confidence diagnosis may still be unreliable, as uncertainty estimation reveals hidden fragility in predictions. Medical AI systems must distinguish between aleatoric uncertaint…
Stylist Law Roach told Good Morning America on June 29 that AI-generated images of Zendaya's wedding dress were inferior to the real gown, saying 'That dress was not good enough' and 'Trust me, the dress is better than t…
California Governor Gavin Newsom announced a partnership with Anthropic to provide state agencies, cities, and counties with discounted access to the Claude AI assistant at a 50% discount, along with free workforce train…
CORE, an open-source governance runtime for AI-assisted software development, enforces deterministic constitutional rules to block unsafe AI actions before execution. The tool, available on PyPI, provides auditable conse…
Anthropic has partnered with California to provide state and local government agencies with a 50% discount on its Claude AI tools, free workforce training, and technical support. Governor Gavin Newsom announced the deal …
Taiwanese prosecutors raided Supermicro's office and 11 other locations on June 29 as part of an investigation into an alleged $2.5 billion scheme to smuggle Nvidia's restricted AI chips to China. The operation, led by t…
Anthropic lobbied for US government restrictions on AI models like Mythos, Fable, and GPT-5.6, leading to a system where only wealthy companies can access the best models. The author argues for freedom of intelligence, o…
Qihoo 360 founder Zhou Hongyi unveiled Tulong Feng, an AI vulnerability agent, at ISC.AI 2026, claiming it as China's answer to Anthropic's Mythos. Separately, Z.ai released GLM-5.2 under an MIT license, which outperform…
Venture capitalists Marc Andreessen and Ben Horowitz backed Donald Trump's 2024 campaign claiming he would protect AI, but Trump is now shutting down AI initiatives. The VCs' real motivation was access and power, as they…
TiDB completed a 45-day white-box security assessment by NCC Group, which found zero critical, high, or medium severity issues across 20 repositories. All eight findings were rated Low, covering default configuration har…
Cybersecurity teams are struggling to keep up with rapid AI adoption as employees use thousands of AI tools without organizational oversight, creating data security blind spots. Fortinet executive Robert May warns that s…
Senator Mark Warner is preparing a discussion draft bill that would impose a "duty of loyalty" on AI agents, requiring them to prioritize users' interests over those of developers, platforms, or advertisers. The proposal…
Former Bayhill High School teacher Jason Hoopes, 50, was sentenced to six years and eight months in prison after pleading no contest to arranging a sexual meeting with a minor and statutory rape. Hoopes groomed a student…