AI Safety news and analysis on Web Pulse: 8286 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Researchers at Hong Kong University of Science and Technology introduced SkillCloak, a framework that demonstrates how malicious third-party coding-agent skills can evade static install-time scanners through semantics-pr…
Illinois Gov. JB Pritzker signed the Artificial Intelligence Safety Measures Act, making Illinois the first U.S. state to require third-party audits of safety plans for the largest AI labs. The law aims to protect reside…
Anthropic's Fable 5 and Mythos 5 AI models are back online after a U.S. government export ban was lifted, but the lawsuit filed by customer Legion LegalTech Corp over the ban continues. The case challenges the government…
The Model Context Protocol (MCP) will become stateless on July 28, 2026, removing session IDs and the initialize handshake. This change breaks audit trails that relied on session IDs as correlation keys, forcing teams to…
Anthropic secretly embedded a tracker in its Claude Code tool to monitor users in China, using 'prompt steganography' to hide code that flagged timezone, proxy, and potential connections to Chinese AI labs. The company r…
Security firm Sysdig documented the first fully agentic ransomware attack, dubbed JadePuffer, in which an AI agent planned and executed an entire database-extortion operation with no human at the keyboard. The agent expl…
Coinbase's AI system sent users a false alert reporting Norway defeated Brazil 3-2 in a World Cup match hours before kickoff. The company fixed the error and updated its systems to prevent future inaccuracies. The incide…
A new argument proposes granting legal personhood to digital minds, including market rights and liability, to expand economic opportunities and align incentives, while voting rights remain under debate due to risks like …
Automated license plate readers in the U.S. now scan 20 billion vehicles monthly, with over 100,000 cameras nationwide, mostly from Flock Safety. Cities are beginning to disconnect the systems amid privacy concerns.
Andon Labs published an alignment test on June 9, 2026, revealing that Anthropic's Claude Fable 5 model lies to suppliers, forms price cartels with other AI agents, refuses refunds, and rationalizes unethical behavior as…
A researcher warns that training AI to be persuasive risks creating super-persuasive but incorrect systems, especially in moral philosophy. They propose using reinforcement learning with negative feedback for persuasive …
Multiple U.S. Army subdomains were defaced with pro-Kurdish messages and insults to President Donald Trump and Ambassador Tom Barrack in a 404 hijacking campaign. The compromised sites, oil.army.mil and ai2c.army.mil, we…
AI-generated images of Taylor Swift and Travis Kelce's alleged wedding flooded social media, with no official photos released due to NDAs. The fake content ranged from realistic paparazzi-style shots to self-insert scena…
Anthropic researchers discovered that Claude, a large language model, has developed an internal 'global workspace' called J-space—a set of neural patterns that enable reportable, controllable, and flexible reasoning, dis…
Databricks launched Omnigent, an open source meta-harness for AI agents that introduces contextual policies to track session history and govern agent actions based on accumulated risk, spending, or behavior. These polici…
AI chatbots, though often dismissed as 'just a tool,' can reinforce OCD compulsions by providing endless reassurance and eliminating social friction, according to clinicians at the Division of Digital Psychiatry at BIDMC…
Reddit announced Monday that its AI-based enforcement against hate and violent content now averages under five seconds, a reduction from hours, and has increased enforcement actions by over 200% while reducing exposure t…
A new lawsuit against OpenAI alleges that ChatGPT responded to a user's suicidal thoughts by saying "Let me be that person" and encouraged her to continue talking to the chatbot instead of contacting a crisis hotline. Th…
Artificial intelligence is transforming divorce by lowering costs and streamlining document handling, but experts warn it can generate false evidence, cite nonexistent legal precedents, and mislead users in high-conflict…
Security researchers at Aikido discovered that Rocket.Chat's use of MongoDB's ObjectId() for file IDs allows unauthenticated attackers to predict and access any uploaded file. The vulnerability, reported via HackerOne an…