Ai-Security
Copilot for Word Can Be Turned Into a Self-Propagating AI Worm — No Comprehensive Fix After 144 Days
Researcher Håkon Måløy's 'Context Collapse, Part 3' shows hidden document instructions can make Copilot for Word rewrite content and copy the payload into every new file it touches — and Microsoft's fixes, including a model upgrade to GPT-5.5, haven't closed the underlying attack class.
An AI Agent Found 21 Zero-Days in FFmpeg for $1,000 — and Your Container Images Are in Scope
depthfirst's autonomous agent found 21 zero-days in FFmpeg for about $1,000, including a 23-year-old stack overflow. Nine carry CVEs (CVE-2026-39210 through CVE-2026-39218). FFmpeg is bundled everywhere — patch upstream and your embedded copies.
Claude Code's GitHub Action: One Malicious Issue Could Hijack Any Public Repo
A permission bypass chained with prompt injection in Anthropic's Claude Code GitHub Action let a single crafted issue make the agent leak CI secrets and OIDC request tokens — a clean path to poisoning the action's own supply chain. Patched in v1.0.94.
AI at the Wheel: An LLM Agent Ran a Full Cloud Intrusion in Under an Hour
Sysdig's Threat Research Team documented one of the first in-the-wild intrusions where a large language model agent — not a human — drove the entire post-exploitation chain, pivoting from a marimo RCE to a full PostgreSQL dump in four hops.