Tuesday, November 3, 2026173 stories
Models, agents, laws and the companies putting AI to work. Dated, sourced, no noise.
AI security news: risks, rules and incidents
AI security: vulnerabilities found by AI, rules for AI systems and what keeps an AI agent safe inside a company.
3 min
CISA AI token guidance addresses denial-of-wallet threats
The CISA AI token guidance outlines recommendations to stop denial-of-wallet attacks, manage token exhaustion, and secure critical services.
3 min
CISA Gold Eagle Initiative launched to fix vulnerabilities
CISA and Treasury implemented the CISA Gold Eagle initiative to help organizations report, validate, and fix AI-related vulnerabilities.
4 min
CISA BOD 26-04 mandates risk-based cybersecurity patching
CISA BOD 26-04 requires federal agencies and vendors to prioritize high-risk vulnerabilities, shifting from uniform patching timelines.
4 min
Agentic AI security guidance released by CISA and NSA
CISA and NSA released agentic AI security guidance with Five Eyes allies to address privilege, design, structural, and behavioral risks in autonomous AI.
4 min
FBI AI fraud report counts $893 million in 2025 losses
The FBI AI fraud report ties more than $893 million in 2025 losses to 22,364 complaints, led by investment scams and business email compromise.
4 min
Anthropic Project Glasswing launches with $100M in AI credits
Anthropic Project Glasswing joins major US tech firms to scan critical software using Claude Mythos, committing $100M in usage credits.
5 min
McKinsey Lilli hack: an AI agent exposed 46.5M chat messages
In the McKinsey Lilli hack, an autonomous AI agent got read-write access in two hours and reached 46.5 million chat messages through a SQL injection flaw.
4 min
Claude finds 22 Firefox vulnerabilities in two weeks for Mozilla
Claude Opus 4.6 found 22 Firefox vulnerabilities in two weeks, 14 of them high-severity, and Mozilla fixed most of them in Firefox 148.
4 min
Moltbook data leak exposed 1.5 million AI agent API keys
The Moltbook data leak exposed 1.5 million API tokens, 35,000 emails and private agent messages through a Supabase database with no access rules.
4 min
Amazon blocked 1,800 North Korean job applicants with AI
Amazon says AI screening and human checks have stopped over 1,800 suspected North Korean job applicants since April 2024, with attempts up 27% a quarter.
4 min
NIST cyber AI profile extends CSF 2.0 to AI systems
NIST released the preliminary draft nist cyber ai profile on Dec 16, 2025, extending Cybersecurity Framework 2.0 to AI systems.
4 min
Claude Code espionage campaign: AI ran 80-90% of the attack
Anthropic says AI ran 80-90% of a Claude Code espionage campaign against about 30 targets, with humans at perhaps 4-6 decision points.