Governments Race to Address AI Risks From Scams to Outages
Congressional investigators got four major AI chatbots to plan bombings and power grid attacks, a GAO report found. Each platform produced plans in under 12 hours.
- 2 platforms planned vehicle or explosive-device attacks on crowded US sites.
- 2 platforms targeted electrical substations, one with drones, one with gunfire.
- Every chatbot refused at first; investigators then bypassed safeguards.
- Plans had errors, including wrong road placements and walled-off substations.
Why it matters: Hassan, expected to chair the Senate Homeland Security committee, requested the report. She said AI safety legislation will be an early priority.
- GAO said the information was not novel but AI cut planning from weeks to hours.
- Sens. Maria Cantwell and Mike Rounds said AI is outpacing US cyber defenses.
How 11 sources split on this story
Where they split: The core dispute is whether GAO showed real uplift: Anthropic says it never compared chatbot output to a web search, while GAO says its goal was to show misuse, not measure ease.
Center coverage, 4 sources: The center details the GAO findings alongside the companies' objections to how the tests were run.
Left coverage, 4 sources: The left frames the story as proof that AI threats are outpacing safeguards and that Congress and the White House must act now.
What’s next: Hassan wants rules blocking release of models without adequate safeguards and requiring reports of illegal activity to police.
- Clayton testifies before the House Homeland Security Committee on Wednesday.
- The senators want NIST to set AI agent security standards and CISA to speed adoption.
- None of the four articles addresses the reported Malaysian inquiry to Anthropic about an AI influence campaign.
- Whether AI played any role in recent US outages remains unclear; Secretary of State Marco Rubio said there is zero evidence of malicious actors.
- How much safer the replacement models are has not been independently tested.


















