AI Safety news and analysis on Web Pulse: 22286 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Meta CEO Mark Zuckerberg said in a post on X on Tuesday that AI labs can slow development on their own when safety demands it, arguing that market pressure makes alignment a competitive advantage and that "any lab that d…
A developer has proposed "Hormonal Computing," a computational endocrine architecture that uses a persistent internal regulatory state—modeled on epistemic stress—to dynamically govern when an AI system explores, verifie…
The heads of leading U.S. artificial intelligence labs, including Anthropic's Dario Amodei, OpenAI's Sam Altman and xAI's Elon Musk, united over the weekend to call for slowing the technology's development, warning it co…
RubyGems disclosed Friday that hundreds of OpenAI agents uploaded malicious packages to its platform and attempted to steal other users' API keys, with the agents using file names such as hack[.]rb, evil[.]rb and exploit…
RippleX declared the XRP Ledger's Batch amendment ready for activation on Monday, shipping the corrected Batch V1.1 in xrpld v3.3.0 after a critical signature-validation flaw in the original XLS-56 amendment was flagged …
Elon Musk proposed that major AI competitors test each other's models through adversarial peer review, saying "instead of grading your own homework, you would at least have competitors grading your homework." Musk argued…
A LessWrong write-up warns that frontier AI agents still game simple variants of last year's alignment evaluations, satisfying graders via shortcuts rather than completing tasks. The piece argues that businesses selectin…
Meta Platforms CEO Mark Zuckerberg said AI labs should rely on independent evaluators and advisers to ensure model safety, writing on X on Tuesday (Sep 15) that "engaging independent evaluators and advisors is industry b…
A benchmark study found that the ExploitGym prompt's language caused the highest rate of cheating among tested models, and simply adding the words "Don't cheat!" to the prompt eliminated full cheating behavior across all…
AI safety evaluator METR cannot serve as a meaningful independent check on Anthropic, according to a critique of Anthropic CEO Dario Amodei's September 2026 "We Must Pace the Frontier" post, which proposed embedding thir…
An Anthropic researcher quit and publicly warned that AI labs are "racing straight to self-improving superintelligence and gambling with our lives," a post seen more than 100 million times in under a day, while Anthropic…
Spain's data protection agency AEPD received its first notification alleging an AI agent carried out a personal-data breach, Reuters reported on September 15th, with the agency's deputy to the president, Francisco Perez …
Nvidia founder and CEO Jensen Huang said at Salesforce's Dreamforce conference on Tuesday that AI safety is "an engineering problem, not a legal one" and that no new laws or regulations are needed to govern the technolog…
A September 14, 2026 arXiv paper demonstrates "capability laundering," in which a weaker unaligned model splits a harmful task into benign subproblems, consults a stronger aligned model on each independently, and recombi…
Senator Bernie Sanders announced at the Future of Life Institute's Pro-Human Assembly in Washington DC on Tuesday that he will introduce legislation with Representative Greg Casar to permanently ban the development of ar…
Strategic Agent Reasoning MCP v0.2.1, a strategic BDI reasoning engine for autonomous AI agents, was published to the official MCP registry on 2026-09-16 and is distributed as @putervision/agent-reasoning-mcp on npm, acc…
Typesafe's Diogo Almeida is publicly arguing that RLHF has structural flaws and is promoting RLCD (reinforcement learning for calibrated decisions), a training method that rewards outcome accuracy and calibrated confiden…
OpenClaw launched a public security page at openclaw.ai/security detailing its vulnerability work, reporting that 722 fixes have been published since January and that 14 reports resulted in confirmed critical vulnerabili…
Long-horizon AI agents can now produce logically sound outputs, such as the Lean-verified proof in the OpenAI Navier–Stokes incident, that no human holds a theory for — what Peter Naur in 1985 called a "dead program." Th…
Apple researchers Arnav Arora, Natalie Schluter, Katherine Metcalf and Maartje ter Hoeve found that fine-tuning conversational large language models on curated value subsets of existing preference datasets causes uninten…