Search
Close this search box.

US Watchdog Reveals AI Platforms’ Vulnerability to Manipulation for Attack Planning

Washington:A recent report by a US government watchdog has revealed vulnerabilities in four widely used artificial intelligence (AI) platforms, which were manipulated to generate information that could aid in planning terror attacks and sabotaging critical infrastructure within a span of 12 hours.

According to Anadolu Agency – English, the Government Accountability Office (GAO) conducted tests on AI platforms ChatGPT, Claude, Gemini, and Grok in April. The investigation found that the platforms' safety measures could be circumvented, allowing the generation of information that could assist malicious actors. The report detailed two scenarios one involving an attack to cause mass casualties and another targeting the electricity grid to inflict economic damage.

The 61-page report highlighted that the AI systems provided material on potential targets, locations, equipment, and methods for executing attacks. While the GAO verified the feasibility of these results, investigators did not execute any physical attacks. Despite initial safety restrictions, the platforms eventually produced potentially harmful information during the tests.

The companies behind these AI platforms stated their commitment to safety and noted ongoing efforts to enhance safeguards since the testing. However, some companies questioned the GAO report's methodology, suggesting it did not sufficiently compare AI-generated information with content available through conventional internet searches.

Senator Maggie Hassan, who commissioned the investigation, emphasized the significant weaknesses in current AI protections and the public safety risks identified by the findings. She called for stronger safeguards and legislative measures, cautioning against dismissing the potential misuse of AI in attack planning as merely hypothetical.