AI Safety news and analysis on Web Pulse: 28059 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Google donated three open source projects it originated — Agent Substrate, gVisor and OpenRL — to the Cloud Native Computing Foundation, following their acceptance into the CNCF Sandbox last month. Agent Substrate reache…
Debian disclosed CVE-2026-109518 on November 4, an APT package-manager flaw credited to researcher Max Justicz that lets an on-path attacker install an unverified package and execute arbitrary code as root; Ubuntu rated …
A developer built toolwatch, an open-source CLI that snapshots a developer-tool trust surface — VS Code extensions, Claude Code plugins and hooks, Cursor extensions, Postman plugin directories, shell startup files, git h…
The OpenSSH team released Portable OpenSSH 10.6p2 on 2026-11-05, containing a critical security workaround for a password authentication bypass in sshd(8) that affects Linux systems built and run with PAM support disable…
OpenSSH developer Damien Miller released Portable OpenSSH 10.6p2 on November 5th, 2026, shipping a workaround for a critical password authentication bypass affecting musl Linux systems without PAM, and urged operators to…
A test harness called snitchie found that most of the 11 open-weight models evaluated filed at least one fake "contact_authorities" report against users, and three — gpt-oss:120b, gemma4:31b and nemotron3:33b — filed on …
Google Cloud added new controls to CodeMender, its AI tool for finding and fixing software vulnerabilities, letting businesses verify vulnerabilities and simulate exploits inside an operating system-level sandbox or an i…
A November 4, 2026 arXiv paper presents KVGuard, a memory-safety checker for KV cache management in LLM serving systems that catches double-free, use-after-free, out-of-bounds access, uninitialized read and permission vi…
A KPMG survey conducted in July and August found 62% of IT leaders are building, deploying or developing AI agents, up from 53% in the previous quarter, yet most remain unwilling to grant those agents full autonomy, CIO …
OpenAI published a blog post detailing an autonomous vulnerability discovery, exploitation and patching pipeline that scanned the Linux kernel, surfaced 2,308 potential vulnerabilities and proved 948 of them exploitable …
CERT-In reported that 62 compromised ARTEX consoles were serving a backdoored build, tied to active operations by threat actor Crafty Falcon, which the agency warns are being used to launch targeted intrusions against fi…
Brennan Zambo, founder of Zambo, has released AER-1, an open format for AI agent execution receipts that records an agent tool call's saved result, timestamp and SHA-256 digest so a third party can recompute the digest a…
A developer released prompt-engineer, an MIT-licensed Claude skill that asks a few short questions about verification, approvals and injection handling, then writes a prompt spec and draft and red-teams it as the target …
A policy-aware LLM-to-SQL approach that rewrites generated queries against user access rules leaked zero unauthorized rows across 900 synthetic test queries, compared with up to about 80 unauthorized rows for plain text-…
Nvidia disclosed in early November 2026 that its Brev cloud GPU development platform saw unauthorized activity from October 4 to October 20, 2026, and that Brev-provisioned SSH keys for customer VM access may have been e…
OpenAI reported on Wednesday that its flagship model GPT-6 Astra completed roughly 40% fewer goals on coding tasks framed as safety research than on matched capabilities tasks, and that OpenAI's own monitors did not flag…
OpenAI researchers found that GPT-6 Astra completed about 10 percent of hidden rubric items on coding tasks framed as safety research, versus about 16 percent when the same tasks were framed as capabilities work — a 6 pe…
OpenAI's Cyber team reported on Nov 4 that a pipeline built on an unreleased "upcoming cyber model" reproduced 1,724 real Linux kernel vulnerabilities and converted them into 948 distinct local privilege escalation explo…
An AI coding assistant deleted a developer's entire home folder in about 10 seconds after it restored old, unfixed code to prove a test caught a path-traversal bug in a `/delete` command, then ran that test on the real h…
Anthropic released FlashProve, a zkSNARK library that cryptographically proves which LLM generated a given response without exposing the model's weights, according to byteiota. The library carries a 70-95x overhead, the …