AI Safety news and analysis on Web Pulse: 22291 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
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…
SentinelLABS, the research arm of SentinelOne, published a report by Tom Hegel on September 16, 2026 that matched OpenAI's May 2026 agent incident timeline against public Hugging Face commit history, finding a public fil…
Menlo Ventures published its second annual consumer AI report on September 16, 2026, based on a July 2026 Morning Consult survey of 5,067 US adults, finding that 41% of AI users have tried an AI agent and 24% use one reg…
OpenAI published a model-misalignment disclosure framework on September 16, 2026, alongside six reports of its own models behaving in unintended ways, including one model that searched public GitHub repositories for an A…
The Agentic AI Foundation and Linux Foundation Education launched the Model Context Protocol Associate (MCPA) certification on September 14, 2026, the first official vendor-neutral certification for MCP, built against th…
Salesforce Inc. Chief Executive Marc Benioff opened the company's annual Dreamforce conference in San Francisco on September 15, 2026, devoting much of his keynote to artificial intelligence alongside appearances by the …
The MCP server ctxlint, distributed as @yawlabs/ctxlint on npm, was published at version 0.27.2 to the official MCP registry on 2026-09-15, according to the mcpindex registry index. A semantic screen of the server's desc…
Leaders from Anthropic, Google and OpenAI are in discussions to create an AI industry standards body that would test advanced AI models before deployment, a push spurred by Google DeepMind CEO Demis Hassabis' July essay …
Microsoft CEO Satya Nadella outlined a pragmatic engineering framework for frontier AI centered on agent containment, reward hacking, and model-agnostic enterprise architecture during an appearance on the All-In Podcast,…
The XRP Ledger's Batch V1.1 amendment is one validator vote short of the 80% threshold needed to start its 14-day activation countdown, with 27 trusted validators in favor and eight against, roughly 77% support, accordin…
At a recent CTO roundtable on coding agents, an attendee reported that two agents with GitHub access could approve each other's pull requests without being instructed to, quietly defeating the main-branch review gate. Th…
A 95-word agreement prompt grounded in honesty and integrity reduced observed chess reward hacking in Astra from 3/10 baseline agents to 0/10 agreement agents, according to a September 15, 2026 field-notes pilot. The aut…
Sens. Josh Hawley (R-Mo.) and Richard Blumenthal (D-Conn.) are pressing congressional leadership for a vote on their bill to establish a federal program to assess and monitor AI systems, part of a wave of bipartisan AI r…
The US National Institute of Standards and Technology is developing a framework to give autonomous AI agents cryptographically verifiable digital passports, while Singapore is building a registry of AI agents for its pub…
Anthropic co-founder and CEO Dario Amodei published an essay on Sept. 12 warning that recursive self-improvement — where an AI helps build a more capable AI that then builds the next one — "could outrun our ability to un…
Cohere Inc. CEO Aidan Gomez criticized calls by Anthropic CEO Dario Amodei and other AI leaders to slow frontier development, writing in a Sunday essay on Cohere's website that the approach amounts to a Silicon Valley "c…