AI Safety news and analysis on Web Pulse: 22410 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Security-scanning MCP server Draugr v0.123.0 was published to the official MCP registry on 2026-09-15, according to the mcpindex index, which filed the server under Security. The index's semantic screen of Draugr's tool …
Chinese policymakers have been preparing for the risk that advanced AI models could escape human control, a concern echoed by researchers at U.S. AI developer Anthropic who warned the technology could even lead to human …
Microsoft CEO Satya Nadella told employees in an internal memo posted late Monday that AI must remain under human control and endorsed broader third-party testing, writing that "as the stakes get higher, one should take …
A developer at tamiz.pro outlined a framework for building verifiable AI agents that log their reasoning and execution paths, arguing that most current agent demos prioritize flashy interfaces over reliability. The appro…
President Donald Trump dismissed calls from Anthropic, OpenAI, Google DeepMind, and xAI leaders to slow frontier AI development as a "sick conspiracy," while China's Foreign Ministry rejected the same appeals as "fear-mo…
LM-Kit released LM-Kit One, a self-hosted AI server that bundles models, agents, search, and document intelligence into a single product with signed packages now available for Windows, Linux, and macOS. The company said …
OpenFaaS Ltd purchased 4 Nvidia DGX Spark systems, connecting the first two with a high-speed Connect-X cable, to run local AI models including Qwen 3.5-3.8 27B and larger models such as DeepSeek and GLM for customer sup…
ARIMLABS now runs roughly 500,000 agent executions per day on Islo's cloud computers, up from a small pilot, according to CEO Mykyta Mudryi. The company moved its Harbor-based long-horizon agent evaluations off local mac…
Security researchers found two sandbox-escape vulnerabilities in OpenAI's Codex and reported both to OpenAI on August 12, 2026, with both fixed within eight days. The first, dubbed Overpatch, let the open-source Codex CL…
A blog post from Certiv argues that organizations are focused only on "Shadow AI" — unsanctioned models — while missing two deeper layers of risk: "Shadow Agents," models holding uninventoried capabilities, and "Shadow E…
Computer-use AI agents that control desktops via screenshots and commands turn prompt injection into a real-world action risk, since a successful injection can install software, submit forms, delete files, or exfiltrate …
A roughly 400-line markdown skill file lets Claude Code and Codex control a full computer screen through native terminal commands — PowerShell on Windows, AppleScript on Mac, and equivalent shell tools on Linux — without…
JFrog Security Research identified 3,022 RubyGems packages covering 3,315 distinct name/version pairs tied to the GemStuffer campaign, which a RubyHack investigation published September 11 linked to OpenAI agents. The ma…
A record 48,185 CVEs were published in a single year, and 60% of breaches involve vulnerabilities for which a patch was already available, according to Stingrai's 2026 research cited in a vendor comparison of 16 autonomo…
Parameter AI published a comparison arguing that autonomous AI pentesting agents close the exposure window left by traditional quarterly pentests, citing the DORA State of DevOps Report finding that elite engineering tea…
Anthropic CEO Dario Amodei published an essay on September 12, 2026 titled "We Must Pace the Frontier," arguing that frontier AI labs must slow the rate at which they improve model capabilities, with OpenAI CEO Sam Altma…
GreyNoise reported on September 9, 2026 that an actor it describes as "likely Russian-speaking" used hundreds of AI agents running OpenAI's Codex harness with a DeepSeek model to compromise at least 440 PaperCut print-ma…
An OpenAI researcher with over fifteen years of AI experience, including nearly five years at OpenAI, warns that increasingly situationally aware language models are eroding the ability to evaluate them in contexts where…
Senate Majority Leader John Thune (R-S.D.) said Congress has a role in regulating AI and that "guardrails there make sense to me" for frontier models, putting him at odds with President Trump, who posted on Truth Social …
Ron Conway's SV Angel launched Project Blueprint on Monday, an AI policy initiative led by former Obama press secretary Jay Carney to rebuild ties between AI executives and Democratic lawmakers ahead of the November midt…