AI Safety news and analysis on Web Pulse: 11310 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Neuroscientists from Université de Montréal and Johns Hopkins University warn that people risk mistaking AI intelligence for actual consciousness, noting that advanced chatbots like ChatGPT and Claude can appear fluent a…
New research from Kapwing reveals that 59% of videos served to a new TikTok user's For You page are AI slop, three times the rate found on YouTube. The study also found that 57.4% of TikTok videos aimed at children are A…
Norway has implemented a near-total ban on artificial intelligence in elementary schools, effective 2026, citing concerns over data privacy and child development. The policy restricts AI tools in classrooms for students …
The US government forced Anthropic to pull its Fable 5 and Mythos 5 AI models over national security concerns after Amazon researchers found a way to bypass Fable 5's guardrails. Cybersecurity researchers signed an open …
Andrea Volpini argues that ontologies are becoming the control plane for AI agents, while Sergey Vasiliev warns that scaling unvalidated meaning creates ontology debt. A 2025 systematic review confirms the field overinve…
A developer compiled a chronological review of public AI-agent incidents from the past sixteen months, arguing that the root causes are not model failures but mundane engineering issues like unconstrained permission scop…
A developer is building 'doll', an open-source personal AI continuity system designed to preserve user state independently of models, providers, or applications. The system aims to ensure that personal AI environments re…
OpenAI hired Dean Ball, a former White House AI adviser who helped write the Trump administration's AI Action Plan, to lead a new policy team called Strategic Futures, and also recruited Noam Shazeer, a co-inventor of th…
The Trump administration ordered Anthropic to obtain US approval for foreign nationals to use its AI models, marking an unprecedented use of export control laws that raises legal questions about government authority over…
A developer warns that shipping an AI agent without a kill switch is a liability, recounting a finance bot that issued refunds in a loop due to the lack of a stop mechanism. The developer argues that building the agent i…
The AI industry's focus on model intelligence is becoming irrelevant as trust emerges as the true bottleneck. Adoption fails not due to capability but human willingness to hand over decisions, and national investments in…
The upcoming MCP 2026-07-28 specification introduces breaking changes including a stateless protocol, removal of the initialize handshake, new transport headers, and stricter security requirements. Server operators must …
Five reputed gang members were sentenced on June 18, 2025, for the 2023 murders of three people at scenic lookouts in Pasadena and Rancho Palos Verdes. Four received life without parole, and one got 30 years to life. The…
Sen. Marsha Blackburn is negotiating with the White House on a bill package that would trade child safety and deepfake protections for preemption over state AI laws. The Fable incident has spurred Congress and the G7 to …
Pasadena Police Chief Gene Harris told a city council committee that an investigation into a September officer-on-officer shooting that seriously injured one officer is examining whether it was part of 'chronic behavior'…
Holotta Tyme, who bought Club 1220 in 2020, keeps the only LGBTQ bar in Contra Costa County alive in Walnut Creek, preserving a historic safe space that opened in 1976. The bar has survived discrimination, epidemics, and…
Dapr has introduced cryptographic tamper detection for workflow execution histories by signing each history event with the sidecar's mTLS identity, creating an auditable hash chain. The feature requires careful root CA l…
Google, Microsoft, OpenAI, and other tech companies launched the Appia Foundation to create modular specifications that help enterprises demonstrate their AI systems comply with global standards and regulations. Hosted b…
Sovereign AI is shifting from a software model race to a supply chain security issue, as governments and companies worry about access to frontier AI models. The real challenge lies in securing the entire AI infrastructur…
Relatives of Bryan Omar Roa, one of five men accused of plotting to attack a UFC event at the White House, told the FBI they noticed his worsening behavior but did not report him to authorities, according to a criminal c…