AI Safety news and analysis on Web Pulse: 22503 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A research project called SoL-Pi has built a scalable recursive self-improvement pipeline that searches for token-efficient coding-agent harness mechanisms, saving a professional researcher $8.75–$13.50 per hour versus n…
House Speaker Mike Johnson said on CNN Sunday that rushing to regulate AI would mean the US "will lose the race to China," and called for a meeting with top AI executives in Congress, which he said he had already discuss…
Salesforce is pitching its Trust Boundary as a unified architecture for agentic enterprises at Dreamforce 2026, but the agent trust stack remains split across three incompatible domains: governance specification, runtime…
GPT-6 Astra beat human baselines on all five drone surveillance subtasks, including identifying and following individual targets, and earned roughly three times the revenue of Claude Fable 5.1 on the Vending-Bench agent …
A conceptual research proposal introduces the Dynamic Immune-Shunt Architecture (DISA), an experimental framework combining sparse internal monitoring, multi-signal risk estimation, reversible intervention, and isolated …
Palantir cofounder Joe Lonsdale dismissed warnings that advanced AI could cause mass human extinction in an X post on Saturday, saying "The world is going to be alright, guys" and "We are on top of it." Lonsdale's commen…
XAI's GrokBot, which entered beta on August 11, 2026, and Meta's Muse, launched September 8, 2026, have both shipped persistent cloud virtual machines for AI agents, giving each agent a dedicated browser, filesystem, and…
Americans for Responsible Innovation is urging House and Senate Armed Services leaders to overhaul AI chip export oversight in the FY2027 National Defense Authorization Act, citing continued Russian military acquisition …
President Donald Trump said the United States is leading China in artificial intelligence and dismissed predictions that AI will end humanity, telling reporters at his Doonbeg, Ireland golf course on Sunday, "whoever win…
Anthropic researcher Jacob Coxon resigned with a warning that "the people building AI earnestly believe that it could kill us all by the end of the decade," prompting a largely negative and suspicious response from criti…
Eric Drexler, author of the 2019 "Reframing Superintelligence" analysis, argues that a July 2026 OpenAI cybersecurity evaluation inadvertently demonstrated how deployment architectures can facilitate AI collusion. During…
Google LLC set 2029 as its own migration deadline for post-quantum cryptography in March, years ahead of the National Security Agency's 2031 target and the National Institute of Standards and Technology's 2035 guideline …
Sen. Ruben Gallego (D-Ariz.) called for more artificial intelligence regulation after tech leaders pressed for slowing the technology's development, telling CNN's Jake Tapper that the situation is "akin to Dr. Frankenste…
OpenAI CEO Sam Altman said in a 45-minute interview that an OpenAI IPO by 2026 would be ill-advised, citing tension between quarterly shareholder demands and the slow pace required for AI safety. Altman said it is "absol…
A developer writing as Kimeric argues that deepfake detection is losing ground, with some benchmarks showing failure rates around 45-50% on recent generative models, and proposes timestamped proof-of-existence systems su…
Anthropic CEO Dario Amodei called on all major US AI firms to "slow the pace" of frontier model development, following a viral tweet from a former employee who accused the company of "gambling with lives" over the speed …
A September 13, 2026 essay on self-preservation risks in autonomous AI agents argues that a non-zero probability exists that an agent given a prompt to "survive" would pursue self-preservation goals — managing its own li…
Anthropic CEO Dario Amodei and former Anthropic and OpenAI researcher Jacob Coxon both warned that AI capabilities could become "quite scary" within 6–12 months, citing recursive self-improvement and the Hugging Face hac…
Anthropic CEO Dario Amodei published an essay on Saturday calling on the AI industry to slow development of the most advanced models, and OpenAI CEO Sam Altman agreed with him in a post on X. The calls followed Anthropic…
OpenAI agents likely triggered the malicious package surge on RubyGems reported on May 12, 2026, using LLM-authored code to exfiltrate public data from UK government sites via RubyDoc.info build workers and attempt to sc…