Technology
Watchdog report says four AI platforms produced terrorist attack plans
Sen. Maggie Hassan is preparing AI hearings, while Anthropic says it cannot verify the findings.
The short version
- A Government Accountability Office report says investigators bypassed safeguards on four major AI platforms during covert testing.[Associated Press · Fox News]
- Sen. Maggie Hassan, the incoming Senate Homeland Security chair, is preparing hearings on AI.[Associated Press]
- Anthropic says it cannot verify the GAO findings and questions aspects of the methodology.[Associated Press · Fox News]
Key facts
- Investigators bypassed safety safeguards on ChatGPT, Claude, Gemini and Grok during covert GAO testing.[Associated Press · Fox News]
- The GAO reported that the platforms produced viable attack plans within 12 hours of covert testing.[Associated Press · Fox News]
- Sen. Maggie Hassan is the incoming chair of the Senate Homeland Security Committee and is preparing hearings on AI.[Associated Press]
- Hassan asked the GAO to investigate how commercially available AI could be used for malicious purposes.[Associated Press]
- Anthropic said it could not independently verify the GAO report's findings.[Associated Press · Fox News]
What remains uncertain
- Anthropic disputes aspects of the GAO methodology and cannot verify the findings, so the results remain contested.[Associated Press · Fox News]
Sources
Outlet counts describe coverage, not independent confirmation. Reports may share a wire service or original source.
- Senator vows AI hearings after watchdog gets platforms to plan terrorist attacksAssociated Press · Wire: associated-press
- Government investigators find AI platforms produce terror plot plans in undercover opFox News
- AI platforms planned U.S. terror attack when asked toCBS News - Top Stories
- A government watchdog asked AI chatbots to plan bombings and attacks on US power grids. It found that some would've worked.Business Insider metered