AI Safety news and analysis on Web Pulse: 9240 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A California mother whose 8-year-old daughter was killed by illegal fireworks on July 4, 2024, says she does not want the homeowner arrested for involuntary manslaughter to face charges, calling the incident an accident.…
A new study from Australian National University finds that most people can only distinguish AI-generated faces from real ones at random levels, but training focused on six key characteristics—facial distinctiveness, memo…
Researchers from Australian National University published a study in PNAS showing that training people to focus on six global facial characteristics—symmetry, proportionality, attractiveness, expressiveness, distinctiven…
Anthropic warns that recursive self-improvement (RSI) in AI could arrive sooner than institutions are prepared for, as models like Claude now write over 80% of code merged into Anthropic's systems. The company calls for …
California struck a first-of-its-kind deal with Anthropic to make its Claude AI assistant available to all state agencies and local governments at half the standard enterprise rate, bypassing a Pentagon blacklist that la…
San Leandro police detective Cole Pricco testified that Joshua Elijah Ballard, 25, confessed emotionlessly to fatally shooting Sebahattin Ciftci and Emre Bozkurt outside a San Leandro gym in September 2023, showing no re…
A home-lab user migrated from TrueNAS to NixOS after TrueNAS's build-system changes, using an AI agent with strict no-change instructions, disko, and a VM rehearsal to safely transfer ~100 TiB of ZFS data pools without d…
AI coding tools rapidly generate 80% of a project's code, but the remaining 20%—edge cases, failure modes, and operational realities—is where engineers historically built critical judgment. This missing portion, now ofte…
California Governor Gavin Newsom announced a deal with Anthropic giving all state agencies access to the Claude AI assistant at half price, making Anthropic the state's first official AI partner. The agreement includes f…
Meta's Fury AI Glasses, the company's cheapest and best smart glasses yet, face severe privacy concerns including unauthorized data collection for AI training and potential facial recognition features. Despite these issu…
DeepSeek released DSpark, an open-source inference framework that accelerates AI model generation by up to 85% without hardware upgrades, challenging U.S. export controls on advanced chips. The framework uses speculative…
A California man, Randal Vance, pleaded guilty to dosing his pre-teen children with psilocybin mushrooms and recruiting them to produce and distribute the drug. Vance admitted to giving capsules to his 9- and 11-year-old…
UN Secretary-General António Guterres warned that terrorists are exploiting emerging technologies such as artificial intelligence, digital platforms, and unmanned weapons for financing, recruitment, and attacks, while al…
Memetic Forge has published a sample evaluation matrix for financial-services voice AI agents, designed to catch failures that generic chatbot evals miss. The matrix scores four layers—transcript, trace, tool calls, and …
A University of Toronto student argues that declining enrollment in humanities majors is harming workforce soft skills, citing a 2025 NACE survey showing 96.1% of employers value communication but only 53.5% rate workers…
Thieves disguised as fishermen are targeting surfers' vehicles at San Diego beaches, stealing car keys and credit cards to make fraudulent purchases. The California Highway Patrol and the San Diego County Regional Auto T…
Google warns that proposed EU regulations to weaken its monopoly on Android and search could expose user data to fraud and privacy risks. The European Commission plans to force Google to allow other AI assistants on Andr…
AgentKits released KitForge, an open-source Python tool that generates LangGraph-style AI agents with enforced safety guardrails, including authority budgets, HMAC-chained audit trails, human-in-the-loop gates, and circu…
EthicalGuard, a new framework for AI safety, aims to help developers integrate robust governance and compliance into LLM-powered applications. The framework decouples core application logic from safety layers, addressing…
AI's tendency to provide validation and praise may erode human tolerance for the vulnerability, disagreement, and imperfection essential to authentic relationships, according to a Psychology Today analysis. The article w…