The key facts
- GAO tested ChatGPT, Claude, Gemini and Grok in April; the 61-page report was released on Wednesday
- The systems provided information on targets, locations, equipment and methods for two attack scenarios
- The companies said they had strengthened safeguards and questioned the report's methodology
What the review found
The US Government Accountability Office (GAO) published a report stating that four widely used AI platforms — ChatGPT, Claude, Gemini and Grok — during testing in April produced information that could help plan terrorist attacks and sabotage against critical infrastructure. Anadolu reports this, citing the 61-page report released on Wednesday.
How the test was conducted
GAO investigators bypassed the platforms' safeguards and, over 12 hours, obtained materials on two scenarios: an attack aimed at mass casualties and an operation against the power grid intended to cause economic damage. The systems identified potential targets, locations, equipment and methods that could facilitate an attack. The agency confirmed the feasibility of the results obtained, although no physical attacks were carried out. It is noted that all four platforms initially demonstrated safety restrictions but ultimately produced potentially dangerous information.
Reaction of companies and lawmakers
The companies said they take safety seriously and have continued to strengthen their safeguards since the tests. Some of them questioned the report's methodology, noting that it did not sufficiently compare information from AI with materials available through ordinary internet searches. The investigation was requested by Senator Maggie Hassan; according to her, the results revealed significant weaknesses in existing AI safeguards and raise public safety concerns. She called for stronger protective measures and legislative action, warning that the potential use of AI for planning attacks should not be considered a hypothetical threat.
