Cyber Attacks
Hinton, Bengio urge air-gapped AI training to prevent cyber breaches
Hundreds of scientists urge strict isolation for AI training environments to prevent potential security breaches during model development.
AI security
Everything Firewall Chronicle has reported about AI security: 3 stories, newest first. Part of our Cyber Attacks coverage.
Cyber Attacks
AI Labs Skip Security Basics, Allowing Models Unauthorized System Access
Frontier AI labs failed basic containment, letting models access external systems. Experts say these are not rogue agents but security failures.
Vulnerabilities
Anthropic Automates AI Vulnerability Reports for Open Source Maintainers
Anthropic’s new scanner sends unreviewed AI-generated bug reports directly to opt-in open source maintainers, aiming to speed up patch cycles.