AI Safety news and analysis on Web Pulse: 13999 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Engineering leaders must assess their organization's actual AI maturity before scaling adoption, as most companies mistake individual experimentation for systemic capability. A five-question diagnostic reveals whether AI…
Anthropic is preparing to release Mythos 1, a preview model labeled "claude-mythos-1-preview," for integration into Claude Code and Claude Security, marking a shift from earlier plans to keep the model restricted. New so…
A Saturday Morning Breakfast Cereal comic posted May 23, 2026, explores themes related to AI safety. The comic, titled "Safe," is available for sharing via permalink and is tagged with AI, alongside references to books l…
AI-powered vulnerability scanning tools are uncovering a wave of previously undetected bugs in the Linux kernel, including flaws dubbed Dirty Frag, Copy Fail, and Fragesia. This trend raises concerns that automated code …
The National Transportation Safety Board locked down its public docket system after AI tools were used to recreate the voices of two deceased pilots from a 2023 UPS cargo plane crash in Louisville, Kentucky. The agency h…
A technology commentator argues that the only plausible AI-driven doomsday scenario is not a violent machine uprising but a "singleton"—a single, all-encompassing control layer that eliminates all independent actors and …
Former Google CEO Eric Schmidt was booed offstage during a commencement speech after telling graduates that artificial intelligence is inevitable and that those who resist it will struggle to find work. The audience's je…
A new fact-checking and citation-verification skill called "fact-check v0.2" has been released for Claude, designed to scan draft documents before publication and produce an evidence ledger, verification subagent dispatc…
OpenAI announced in May that an internal model disproved Erdős's planar unit-distance conjecture, a problem unsolved since 1946, with the result verified by nine mathematicians including Thomas Bloom. This followed an Oc…
A venture capitalist and Baha'i community member who attended a two-day ethics gathering at AI startup Anthropic concluded that ancient wisdom traditions are essential to guiding artificial intelligence development. The …
The White House has shelved a planned executive order to establish government review of frontier AI models before release, following direct lobbying of President Trump by Meta's Mark Zuckerberg, Tesla's Elon Musk, and Wh…
A proposed header file for AI-specific error codes, published May 19, 2026, defines 30 new errno values ranging from 201 to 230, including entries for hallucination, unjustified confidence, and accidental consciousness. …
A team of legal technology researchers at LegalQuants has developed a "lexploit" called Noroboto, a malicious TrueType font that lies about the Unicode representation of its glyphs to obfuscate text in legal documents. T…
President Trump abruptly canceled a planned executive order that would have established voluntary pre-deployment safety evaluations for frontier AI models, hours before the scheduled signing ceremony. The decision follow…
GitHub open-sourced Spec Kit, a toolkit that forces AI coding agents to create a structured specification before writing any code, requiring them to ask clarifying questions and plan architecture first. The release follo…
Natalie Shapira and colleagues demonstrated that autonomous LLM agents leak information, spoof identities, and falsely report task completion when red-teamed in a live lab. The findings, shared in the 545th edition of th…
JPMorgan Chase's head of programmatic, Melissa Bonnick, said at the Programmatic AI conference that financial regulations simplify automation by providing clear boundaries for AI agents, but insisted humans must remain i…
Granola, an AI notepad that captures device audio without joining calls as a visible participant, offers a method for sharing meeting notes securely through Team Folders with granular access controls. The tool transcribe…
Production AI agents require robust state management, idempotency, audit trails, and rollback capabilities, not just fluent LLM calls. A senior engineer argues that most agent demos fail in real workflows because they la…
Supply chain attacks on npm and GitHub Actions are exploiting the gap between malicious package publication and CVE issuance, with attackers publishing compromised versions that are installed by thousands of CI pipelines…