AI Safety news and analysis on Web Pulse: 12672 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
JPMorgan Chase has blocked its Hong Kong staff from using Anthropic's Claude AI model due to licensing terms, following a similar move by Goldman Sachs. The restriction highlights how geopolitical tensions and US export …
Dream, an Israeli AI cybersecurity startup co-founded by former NSO Group CEO Shalev Hulio, raised $260 million at a $3 billion valuation, nearly tripling its valuation in a year. The company uses AI for predictive threa…
A new study from the Wharton School at UPenn introduces 'cognitive surrender,' where AI users accept AI-generated answers without independent verification. In experiments with 1,372 participants, 73% accepted wrong answe…
AI models used for online hate speech detection vary widely in how they identify and classify hate speech, according to a 2025 University of Pennsylvania study. The study found significant inconsistencies across systems …
Anthropic employees and industry experts allege the Trump administration is targeting the AI company with unreasonable restrictions, including a de facto ban on its latest models, Fable 5 and Mythos 5, citing national se…
A developer argues that when an AI agent's action is blocked by a safety boundary, returning a generic error is insufficient. Instead, the admission layer should provide structured feedback that helps the agent understan…
French President Emmanuel Macron urged the US to grant broader access to Anthropic's AI models, following a US order to block foreign nationals from accessing them. G7 leaders discussed a "trusted partners" scheme to all…
NewCore emerged from stealth on June 15 with $66 million in seed funding to build an identity security layer for AI agents, addressing widespread vulnerabilities where agents use human credentials or shared API keys. The…
The Iran War, which began on February 28, 2026, has become the first major conflict driven by artificial intelligence, with U.S. and Israeli forces using AI to strike over 4,000 targets in four days, while Iran employs d…
Chainguard launched the Athena Coalition, an industry group using AI to find and fix vulnerabilities in open-source software before exploitation. Founding members include BNY, JPMorgan Chase, Cisco, Cloudflare, Docker, K…
A Hugging Face forum user asked about the biggest challenges in AI agent development today, sparking discussion on the topic. The post highlights ongoing difficulties in building robust AI agents, including issues with r…
42Crunch released an API Security Testing Plugin for GitHub Copilot that lets developers find and fix API vulnerabilities within AI-assisted workflows. The plugin addresses growing API security challenges exacerbated by …
A developer shares practical advice for shipping AI agents that survive production, emphasizing that the model is not the bottleneck—state management, crash recovery, and tool integration are. The post recommends LangGra…
A developer compiled a comprehensive defensive knowledge base cataloguing known prompt-injection and jailbreak patterns, the models and systems they have affected, and the defenses against them. The reference, updated th…
Florida Attorney General James Uthmeier filed an 83-page lawsuit against OpenAI and CEO Sam Altman on June 1, alleging product liability and public nuisance over ChatGPT's role in the 2025 Florida State University shooti…
C1 launched enterprise-managed authorization (EMA) for the Model Context Protocol (MCP), enabling governed AI agent access to enterprise tools. Anthropic implemented the first EMA support in Claude, with C1 serving as th…
CLI authentication flows that rely on localhost redirects and browser availability fail in SSH sessions, containers, and WSL environments. The author argues that RFC 8252's OAuth pattern for native apps does not address …
Researchers at an undisclosed institution developed a method to replace attention heads in transformer language models with human-readable Python programs, achieving over 75% similarity in attention patterns on TinyStori…
The UK Home Office knew its AI-powered facial age estimation tool for migrant children was flawed and racially biased, but is rolling it out anyway. A leaked report shows the technology often misjudges teenagers as adult…
A developer describes how to make function-calling in AI agents reliable enough for production, based on experience with an agent that called the wrong tool. Key techniques include keeping tool sets small and distinct, v…