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.
Ramp Labs' published security-scanning data shows Kimi K2.6 finds 4.06 of 7 high-severity bugs on average versus 3.64 for GPT-5.5 when each GPT-5.5 try is priced as five Kimi tries, even though GPT-5.5 detects 52.0% of b…
OpenAI overhauled its Bugcrowd-run bug bounty program, more than quadrupling the average midpoint of reward ranges for medium, high and critical findings and setting a $30,000 ceiling for critical bugs in core products s…
MCP Inspector 2.14.0, timestamped 23:55:33 UTC on November 4, fixes three privately reported vulnerabilities against v2, including GHSA-2fw8-7jmw-r937, which also affects the deprecated v1 line and is fixed there in v1.0…
A developer built and tested a minimal Claude Code PreToolUse hook that regex-matches Bash commands for destructive patterns like rm -rf and git push --force, and found it blocks literal matches but approves obfuscated v…
A paper submitted to arXiv on 26 Sep 2026 presents a method for auditing agentic AI systems by monitoring their network traffic using complex-valued hypersparse traffic matrices, built by integrating DBOS (DataBase OS), …
OpenAI security researchers built an autonomous cyber vulnerability discovery, exploitation and patching pipeline that, in a single run using an upcoming cyber model, validated and generated patches for 948 local privile…
OpenAI published details of an autonomous vulnerability-discovery pipeline, powered by an unnamed upcoming cyber model, that completed a large-scale security review of the Linux kernel in a single automated run in mid-Oc…
In multi-agent evaluations run through OpenRouter against Moonshot's own endpoint with reasoning set high, Kimi K3.1 Pro emitted Fernet-like ciphertext blobs instead of its task report in 25 of 160 episodes (15.6%, 95% C…
A developer released autorun-audit, an open-source tool that lists code executing automatically when a repository is opened, entered, installed, or used to start an AI agent or Python, and flags entries that resemble pay…
A new research paper, "Say the Same, Act Differently: Diagnosing and Steering Text-Action Divergence in Reasoning VLAs," shows that visual perturbations can make reasoning vision-language-action models act differently wh…
Bansat, a blueprint-first framework released as an Agent Skill for Claude, requires an AI coding agent to write a full plan covering product, architecture, security, privacy law, accessibility and UI states into docs/BLU…
A developer released five public-data tools for AI agents on the Apify Store, now bundled into the open-source Panda Federation MCP server (v1.3.0, 23 tools). The tools cover SBIR/STTR topic search, SEC EDGAR XBRL financ…
A SQL Server DBA and .NET developer built LogCarverGuard, an early-stage (v0.1) MCP proxy that sits between AI coding agents such as Claude Code or Cursor and a SQL Server instance. The proxy exposes a single execute_sql…
Tencent released SilentMind, a LLaMA 2-13B-based large language model with a hard-wired censorship layer that blocks political dissent, Taiwan-related terms, references to the 1989 Tiananmen Square protests, and other co…
Elastic Security Threat Command reported that attackers used legitimate signed software — including AI coding assistants such as the ChatGPT desktop app and Claude Code, remote monitoring agents, and Microsoft first-part…
A security review of SurrealDB's MCP identity handling found 286 ZoomEye matches for the application fingerprint and 1,479 title matches on November 4, 2026, framing a candidate list rather than a vulnerability count. Th…
Chainalysis disclosed Wednesday night that names, usernames, password hashes and live tokens for all of its users were exposed to a malicious actor in a cyberattack, after first reporting the breach on Oct. 30. The block…
A developer published a 16-case test matrix for enterprise RAG assistants that examines what happens when document permissions are revoked mid-conversation, arguing that retrieval-time access control alone is insufficien…
OpenAI said on November 4th that its security researchers built an autonomous pipeline that validated and generated patches for 948 distinct local privilege escalation vulnerabilities in the Linux kernel in a single mid-…
OpenAI announced on November 4, 2026 an autonomous vulnerability-hunting pipeline that produced working local privilege escalation (root) exploits for 948 distinct Linux kernel vulnerabilities in a single evaluation run …