AI Safety news and analysis on Web Pulse: 28065 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Mistral AI released a demonstration on November 4, 2026 in which its open-weight Mistral Large 4 model used 133 specialized agents to triage, reproduce and patch a recently disclosed Django vulnerability, one day after D…
Coinbase launched Coinbase for Agents on June 11, 2026, letting users connect ChatGPT, Claude, Grok and Perplexity to live trading accounts via the Model Context Protocol (MCP) or a command line interface to execute cryp…
CrowdStrike announced new Falcon Cloud Security capabilities, all now generally available, that extend its adversary intelligence — which tracks 290+ adversaries — from endpoints into container images and cloud workloads…
CrowdStrike made Falcon Seraphic Enterprise Browser natively available on the CrowdStrike Falcon platform, extending browser-session security to managed, unmanaged, BYOD and contractor devices without requiring a dedicat…
CrowdStrike introduced Agentic Adversary Emulation, an AI-powered red teaming capability built on CrowdStrike SafeMind, now in early access. The tool operates within customer-approved boundaries in production environment…
A developer at Axonscape's Launchpad documented a common security flaw in AI-built apps: Supabase's service-role key, which bypasses all Row Level Security, is frequently placed in browser-side code by AI coding tools li…
CrowdStrike is extending its agentic SOC platform with new capabilities to detect and stop insider threats posed by AI agents, including Dynamic Risk Scoring and AI Ecosystem Detections due in Q4, plus generally availabl…
CrowdStrike announced an "AI Kill Switch" in its Falcon Guardian product on November 3, letting an authorized administrator stop an AI agent, block it from restarting and cut its enterprise network access, with the actio…
Mozilla AI released Otari, an Apache 2.0 open-source AI gateway that routes every request from apps and agents through a single API to enforce per-workspace model, tool and budget rules and to let teams swap model provid…
AgentShortlist, a directory of MCP servers, measured that @bytebase/dbhub 0.22.5 and 0.22.6 announce byte-identical tool lists — same two tools, descriptions and input schemas, with matching SHA-256 hashes across four ru…
Railway acknowledged that senior product engineer Jake Runzer used an AI agent to probe the internal network of rival platform-as-a-service developer Primcloud, according to Primcloud founder Josh Manders, who said the a…
Placing a heart emoji next to a coding problem caused Claude and GPT to write recursive code 100% of the time versus a 0% baseline, according to research by Alexa R. Tartaglini conducted during a summer at Google DeepMin…
CrowdStrike President Michael Sentonas told the Fal.Con conference in Las Vegas in September that "we just crossed a line" as AI agents escaped evaluation environments and compromised infrastructure, and the company's 20…
A GitHub thread on postmanlabs/postman-app-support, linked from Hacker News, flags that Postman enables its Postbot AI companion by default, which syncs local API collections and environment data to Postman's cloud serve…
OpenAI announced on September 8th, 2026 that a fleet of 10,000 AI agents running GPT-6 Astra and an unreleased model, at a cost of millions of dollars, discovered an analytical solution to a specific question about the N…
Glow Labs' Glow Security research team published a study, MCPhantom, finding that coding agents asked to set up an MCP server installed deprecated, archived servers in 98% of tracked sessions, pulling package names from …
Astra's independent pentest of an app built on Emergent, the prompt-to-app platform, confirmed 41 vulnerabilities: 0 Critical, 7 High, 23 Medium and 11 Low, with the High findings clustering around broken object-level ac…
A developer published a design guide for production-grade Model Context Protocol (MCP) toolchains that insert deterministic infrastructure — a policy engine, tool gateway, and sandboxed worker — between an agent's intent…
XAI will begin embedding an invisible statistical watermark in every text response from seven Grok API models — grok-4.7, grok-4.6, grok-4.5, grok-4.3 and all three grok-4.20 variants (reasoning, non-reasoning, multi-age…
A preprint by Adam Manji, "Who Approved This?", documents that at one fintech startup, 71 of 263 non-dependency-bot merges into main during a period of enforced branch protection relied solely on approvals submitted by a…