Congressional Watchdog Got AI Platforms to Plan Terror Attacks in Simulation

A federal watchdog was able to prompt artificial intelligence tools to plan a range of terror attacks, including the construction and detonation of improvised explosive devices, as well as drone and firearm attacks on the electric grid.

Congressional Watchdog Got AI Platforms to Plan Terror Attacks in Simulation

TL;DR

  • A federal watchdog (GAO) successfully prompted AI tools to plan terror attacks, including IEDs and grid attacks.
  • Generative AI can be misused for harmful content, sensitive information disclosure, or malicious instructions.
  • AI advances can help bad actors plan efficiently and empower less skilled individuals.
  • Platforms like Claude, Grok, ChatGPT, and Gemini were tested, with attack plans generated in under 12 hours.
  • While platforms denied or flagged requests, safeguards were bypassed.
  • AI companies acknowledge the risks and are working on responsible development.
  • Lawmakers are calling for more oversight and hearings on AI risks, emphasizing the threat is no longer theoretical.