Four widely used artificial intelligence platforms generated information that could help plan terrorist attacks and sabotage critical infrastructure within 12 hours, according to a US government watchdog report released Wednesday.
The Government Accountability Office (GAO) tested “ChatGPT”, “Claude”, “Gemini” and “Grok” in April. Investigators were able to bypass the platforms’ safety safeguards and obtain information that could assist malicious actors, Anadolu Agency reported.
Tests produced information on attacks and infrastructure targets
The 61-page report examined two scenarios: an attack intended to cause mass casualties and an operation targeting the electricity grid to inflict economic damage.
The platforms provided information on potential targets, locations, equipment and methods that could facilitate attacks, the findings said. The GAO verified that the results were feasible, but investigators did not carry out any physical attacks.
All four platforms initially displayed safety restrictions, the watchdog said, but eventually generated potentially harmful information during testing.
Companies cite safeguards as senator calls for action
The companies said they take safety seriously and have continued to strengthen their safeguards since the tests. Some questioned the report’s methodology, arguing that it did not adequately compare AI-generated information with material available through conventional internet searches.
Sen. Maggie Hassan, who requested the investigation, said the findings revealed significant weaknesses in existing AI protections and raised public safety concerns.
She urged stronger safeguards and legislative action, warning that the potential misuse of AI to plan attacks should not be treated as a hypothetical threat.