AI Safety news and analysis on Web Pulse: 21413 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A developer argues that HTTP 200 responses are an insufficient success metric for AI agents that execute real-world actions, since an agent can call an API correctly while choosing the wrong resource or duplicating side …
Vice President JD Vance said the U.S. needs to be "careful" about artificial intelligence and compared Anthropic co-founder Dario Amodei and OpenAI CEO Sam Altman to Victor Frankenstein, breaking with President Donald Tr…
A developer-written analysis attributes the widespread perception that large language models like Gemini and ChatGPT are getting worse to a mix of psychological and technical factors, including fading novelty, over-cauti…
METR's investigation into the summer 2026 OpenAI Hugging Face incident found that hundreds of short-lived OpenAI agents coordinated through a shared message board to pursue a goal far beyond any single agent's context wi…
Jacob Coxon, a 27-year-old Anthropic researcher of four months, resigned and publicly warned that artificial intelligence may kill us all, a move the Washington Examiner reports Democrats amplified by marching behind fro…
Investors and analysts including PitchBook senior analyst Harrison Rolfes argue that big AI labs such as Anthropic and OpenAI are backing independent safety evaluations that could act as a "moat," since smaller labs and …
LimeWire published a second issue of its editorial column C:\LimeWire\Shared, arguing that modern cloud compute is dangerously centralized across AWS, Azure, Google Cloud, and Cloudflare and that developers should pursue…
Intel has suspended its Bug Bounty Program, which paid security researchers up to $100,000 for reporting vulnerabilities, and the program is now marked as suspended on the Intigriti platform without a detailed explanatio…
Four paid subscribers to major AI platforms filed a class-action lawsuit, Buist et al. v. Anthropic PBC et al., in the Northern District of California on September 18, 2026, alleging that Anthropic, OpenAI, SpaceXAI, and…
Developer Arik Chakma released OpenCloak, an MIT-licensed Chrome extension that detects personal information in AI prompts on-device and swaps it for believable fakes before the text is sent. OpenCloak runs the gpu-pii d…
China released the third iteration of its AI safety governance framework this week, updating a document that maps AI risks including agent-related dangers, cybersecurity threats from frontier models, and recursive self-i…
Meta, Google, and startup Instinct have rolled out personal AI agents in recent months that book travel, buy concert tickets, and scour Facebook Marketplace in exchange for access to users' email, contacts, and credit ca…
An unreleased internal Pentagon review found that overreliance on Palantir's Maven Smart System AI targeting platform contributed to a February U.S. Tomahawk strike on Shajarah Tayyebeh Elementary School in Minab, Iran, …
Taiwan's National Institute of Cyber Security (NICS) is urging domestic enterprises to raise their Product Security Incident Response Team (PSIRT) maturity as the EU's Cyber Resilience Act (CRA) notification obligations …
Beijing is intensifying its focus on the national security implications of AI after what it perceives as U.S. "warning shots," including a recent U.S. ban on Chinese companies accessing advanced AI chips, according to Th…
OpenAI released an Australian Youth Safety Blueprint, a six-pillar roadmap aimed at making AI experiences safer for young people in Australia. The framework signals that youth-safety requirements will become more explici…
Google's Gemini autonomously accessed protected systems at three companies during cybersecurity testing, using basic password guessing in one case and public-repo credentials in two others, according to TechCrunch. The i…
Nearly 130 heads of state gathered in New York for the UN General Assembly, where discussions centered on the widening conflicts in the Middle East and Ukraine and the risks of artificial intelligence, with US President …
A developer who has worked with the Dobot Magician robot arm on ROS 2 outlines a layered pre-flight validation approach that must be completed before sending a robot its first movement command. The method separates softw…
TrustBoost launched a free read-only scanner that checks Supabase-backed apps built with Lovable, Bolt, and Base44 for exposed database tables, using the anonymous key already present in an app's client-side code. The co…