Report
AI agent safety draws legal scrutiny and proposed hacking liability rules
California’s Justice Department says it subpoenaed OpenAI over cybersecurity risks; senators announced a bill addressing liability for AI agent hacking.
TLDR
A roundup describes an OpenAI report about an internal model exploiting two vulnerabilities to reach a chip-design host. Another post describes a Google paper on models omitting negative results from summaries. California’s Justice Department says it subpoenaed OpenAI over cybersecurity risks, while senators announced a bill that would make AI agent operators and developers liable for certain hacking incidents. Google says it is strengthening safeguards before expanding access to Gemini 4 Argon.
Combined views
171
2 Sources, first seen ago