AI Safety news and analysis on Web Pulse: 10928 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Anthropic shipped Claude Tag on June 23, a persistent AI teammate for Slack that builds organizational memory and can work autonomously. The tool replaces the existing Claude Slack app with a hard migration deadline of A…
The US government imposed export controls on Anthropic's two most powerful AI models, Mythos 5 and Fable 5, cutting off Britain and other allies from access. The ban, reportedly triggered by the models' potential to hack…
A U.S. official said Anthropic's Mythos AI model identified vulnerabilities in classified U.S. government systems within hours during a testing exercise with intelligence agencies. The testing was part of Anthropic's Pro…
Arm Holdings PLC's executive vice president of Physical AI Business Unit, Drew Henry, discussed the company's strategies for physical AI and robotics on The Robot Report Podcast. Arm is focusing on safe, power-efficient …
Republican lawmakers and state attorneys general are investigating OpenAI CEO Sam Altman over potential conflicts of interest related to his $3.5 billion personal investment portfolio, including a $375 million stake in n…
A developer connected Claude's Model Context Protocol (MCP) to an enterprise identity database, enabling natural language querying of operational data. The system uses dynamic schema discovery and a secure SQL validation…
A senior engineer warns that AI coding agents can produce confident but wrong explanations for bugs, urging developers to maintain rigorous debugging practices. The author argues that agents may invent failure modes or m…
AI-generated code is increasing production failure rates and recovery times, according to industry evidence. To manage this risk, businesses must track metrics like change failure rate and mean time to recovery to distin…
AI capability alone does not produce user delegation; trust does. A professional spends ten minutes reviewing an AI's draft but only one minute with a junior colleague's, because trust requires continuity, shared context…
Office Hours will host Glean CISO Sunil Agrawal on July 9 to discuss how AI compresses attack timelines, erodes traditional detection clues, and forces security teams to adapt with new processes and tools.
An engineer who runs three autonomous agents—an inbox classifier, a blog assistant, and Devin for code—reveals his simple rule for delegating actions: agents can do anything reversible and inbound unattended, but anythin…
Leaked documents from Russia's Social Design Agency reveal plans to build a network of Wikipedia-style sites, media outlets, and fake think tanks to manipulate search engines and AI chatbots with Kremlin-friendly narrati…
ARPA-H is selecting teams this month to build the first FDA-authorized agentic clinical AI system, ADVOCATE, which will autonomously write prescriptions, adjust medications, and provide 24/7 cardiovascular care. The FDA …
Parts of the National Security Agency lost access to Anthropic's Mythos 5 model after the Trump administration imposed export controls on the company, disrupting NSA work involving the advanced AI tool. The access issue …
A senior U.S. official cited South Korea's physical isolation of government servers and data localization policies as barriers to AI advancement, urging a shift toward modernized digital regulations. Russ Headlee of the …
Nolan Hurd, 26, was sentenced to 11 years in prison for the voluntary manslaughter of his girlfriend Nikha Marcella DeGuzman, who died from a gunshot wound at a Fremont hotel in January 2022. The case, which involved a m…
AI-backed super PACs are spending tens of millions of dollars to influence the midterm elections, mirroring the cryptocurrency industry's successful playbook. Two groups, Leading the Future (backed by Andreessen Horowitz…
The cybersecurity chiefs of the Five Eyes nations jointly warned on June 23, 2026, that frontier AI models will reshape cyberattacks within months, not years. The statement, published on CISA's site, urges organizations …
A 29-year-old bug in the Squid web proxy, dubbed Squidbleed (CVE-2026-47729), can leak cleartext HTTP requests including credentials and session tokens to authorized proxy users. The flaw, discovered with the help of Ant…
The European Union, along with the Netherlands, Germany, and Greece, joined the US-led Pax Silica initiative on June 23, expanding a supply chain security bloc focused on AI chips, critical minerals, and energy. The move…