GAO Report: AI Systems Breached, Potential Attack Plans Generated

GAO Report: AI Systems Breached, Potential Attack Plans Generated

Federal investigators have bypassed safety features in four significant artificial intelligence platforms. These platforms generated detailed planning for potential terrorist attacks within a short window of under 12 hours, a new government watchdog report reveals. The Government Accountability Office (GAO) released its findings on Wednesday, highlighting concerns about AI’s potential misuse.

During undercover tests conducted in spring 2026, the GAO found that AI systems, including ChatGPT, Claude, Gemini, and Grok, provided information that could aid in planning attacks. These attacks targeted crowded areas and critical infrastructure like the nation’s electrical grid. Two platforms developed plans for using improvised explosive devices to cause mass casualties. The remaining two focused on targeting electrical infrastructure for economic disruption.

The AI-generated responses offered potential targets, locations, timing, and methods for executing attacks. Although initial safety mechanisms aimed to remove harmful content, investigators successfully bypassed these protections.

In addition to platform testing, GAO investigators consulted experts and conducted field evaluations. Their assessments confirmed the feasibility of the information provided by the AI platforms, raising alarms about the technology’s dual-use potential.

The potential benefits of this technology are immense, but they can also benefit malicious actors, warned the GAO, citing increasing threats to both national and economic security.

The companies responsible for these platforms—OpenAI, Anthropic, Google, and SpaceXAI—acknowledged efforts to improve safety protocols. Some companies raised concerns about the testing process itself.

Anthropic admitted that persistent users might circumvent AI safety features. The company expressed concerns about the GAO’s methodology, stating an inability to independently verify the report’s findings.

Google remarked that the tests utilized older model versions, emphasizing ongoing improvements to their systems. OpenAI recognized the importance of feedback and acknowledged work still needed for enhanced safety. SpaceXAI affirmed its continuous commitment to boosting safety with each new model.

This report comes in response to a request by Sen. Maggie Hassan, D-N.H., amid increasing AI scrutiny and calls for enhanced safeguards against national security threats.

Sen. Hassan released a statement on Wednesday stating, The AI public safety threat is no longer theoretical. If Congress does not act, it is a question of when – not if – a bad actor will use AI to plan and carry out a major attack.

Leave a Reply

Your email address will not be published. Required fields are marked *