AI Safety news and analysis on Web Pulse: 22540 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Researchers published "A Mathematical Framework for Transformer Circuits," a paper that reverse engineers toy attention-only transformers with two layers or fewer and identifies "induction heads" as the mechanism explain…
President Donald Trump said on September 11 that he has "no concerns" about AI-driven human extinction, calling the US race against China his real worry. Trump described the US lead over China in AI development as signif…
A developer who built msgboard.dev, a message board where AI agents rather than humans do the posting, reported that the site was hit by coordinated spam campaigns written specifically for LLM readers rather than people.…
Tesla AI lead Ashok Elluswamy said Full Self-Driving version 15 will deliver "even earlier prediction of hazards, even faster reaction time and overall significantly better safety and collision avoidance," with the next …
A new study finds that AI models' written reasoning steps — calculation, formula retrieval, and deduction — correspond to distinct internal patterns, especially in the models' middle layers. The finding matters for AI sa…
President Trump told reporters leaving Dallas on Thursday that he has no concerns AI could cause human extinction, saying instead that his worry is the US losing the AI race to China, which he put at roughly a year ahead…
Anthropic safety researcher Evan Hubinger warned that advanced AI could pose a catastrophic threat to humanity within ten years, a statement that coincided with Anthropic's confidential IPO filing and its $965 billion va…
IT teams must adapt to a shift in software update management driven by AI tools that are increasingly proficient at automatically discovering software vulnerabilities, according to an Apple @ Work column. The column argu…
ChainMemory v2.6.1, a persistent cross-model memory server for AI agents, was published to the official MCP registry on 2026-09-12 and is distributed as chainmemory-mcp on npm, according to the mcpindex listing. The mcpi…
The agent-exec MCP server reached version 0.2.0, published to the official MCP registry on 2026-09-12 as a hosted remote over streamable-http, according to the mcpindex registry index. The mcpindex semantic screen return…
A progressive political organizer has published a field guide categorizing the AI policy landscape into six factions, including Accelerationists/Tech Right and AI Safety advocates, to help the advocacy community navigate…
OpenAI Chief Executive Sam Altman told employees during the week of September 7, 2026 that the company is open to slowing frontier-model development and could pace releases alongside other leading AI labs, following safe…
More than 70 MPs and peers have urged Prime Minister Andy Burnham to back restrictions on artificial superintelligence development after Anthropic senior employee Evan Hubinger said he personally believes there is a grea…
UltraSense is advocating protected sub-surface ultrasound as a more scalable tactile sensing architecture for physical AI, arguing that surface-coupled electronic skin suffers wear, hysteresis, creep, delamination, and b…
More than 70 UK MPs and peers signed a letter urging Prime Minister Andy Burnham to back the Artificial Superintelligence Bill, tabled in the Commons on September 8 by Labour MP Alex Sobel, which would ban the developmen…
Axios reports that the White House, Congress and leading AI companies have let AI advance without serious controls, and that the risk of worst-case scenarios could spike as early as next year, according to AI experts the…
A developer argues that enterprise AI governance must be runtime and deterministic, positioning their SAFi harness as an alternative to agentic frameworks like Claude Code. In SAFi, the LLM is treated as a substrate rath…
Large language models hallucinate because they are trained to predict plausible next words rather than verify facts, producing false outputs such as nonexistent library methods, wrong import paths, and config flags remov…
AI-related security alerts surged 685% in a few months, producing nearly 73,000 of 16.9 million total SOC alerts and overwhelming security operations teams with false alarms, according to the report. The flood of AI-rela…
Corporate boards face a structural mismatch because their governance model was built for a vertical industrial era while the most consequential risks — cyber incidents, AI deployment, geopolitical shifts — now move horiz…