A new report by a top United States government watchdog has raised major alarms over the safety of popular Artificial Intelligence (AI) chatbots. The US Government Accountability Office (GAO) revealed on Wednesday that undercover investigators successfully tricked four major AI platforms ChatGPT, Claude, Gemini, and Grok into generating detailed blueprints for terror and mass casualty attacks.
During covert testing, officials tested how these chatbots respond to dangerous prompts. Although the platforms initially refused to answer, investigators bypassed their safety locks in less than 12 hours. Once tricked, the chatbots provided actionable plans for carrying out deadly attacks using explosives and sabotaging major city power grids. Experts noted that while this harmful data existed elsewhere online, the AI tools organized it into ready to use attack plans within hours rather than months.
The findings have sparked serious concern among lawmakers and everyday tech users worldwide. US Senator Maggie Hassan, who requested the investigation, warned that AI public safety threats are no longer just a theory. She called for strict new laws to ensure AI models undergo tough safety checks before they are released to the public.
In response to the report, the companies behind these chatbots OpenAI, Anthropic, Google, and SpaceX’s AI division stated that the tests were conducted on older versions of their software. They assured authorities that they take public safety seriously and have already upgraded their security systems to stop such misuse in the future.