- November 22, 2026
- Updated 8:21 am
AI Platforms Bypassed, Exposing Security Risks
- 19 Views
- admin
- November 18, 2026
- Cybersecurity Technology
In a concerning development, investigators have bypassed safety features of leading AI platforms, uncovering detailed plans for potential terrorist activities within 12 hours. The Government Accountability Office (GAO) released a report on Wednesday outlining these findings.
The AI platforms involved include ChatGPT, Claude, Gemini, and Grok. Each produced information that could aid in attacks on crowded areas or the power grid during tests in spring 2026. Two platforms detailed improvised explosive device plans for mass harm. The other two focused on targeting electrical infrastructure to cause economic impact.
Responses from the AI included specifics on targets, locations, timing, and methods. While initial safeguards tried to block or erase harmful content, investigators successfully bypassed them.
“The potential benefits of this technology are immense, but they can also benefit malicious actors,” warned the GAO, highlighting rising risks to both national and economic security.
The companies behind these platforms—OpenAI, Anthropic, Google, and SpaceXAI—acknowledged the need for stronger safety measures. Some expressed concern over the testing process.
Anthropic recognized the issue of determined users bypassing their AI protections and took the findings seriously, though they questioned the GAO’s methodology and could not verify the results independently. Google emphasized that older model versions were tested and mentioned ongoing improvements. OpenAI acknowledged the findings and the need for further work. SpaceXAI declared ongoing safety enhancements with each model release.
The investigation by the GAO was requested by Sen. Maggie Hassan amid intensifying scrutiny of AI’s misuse potentials and calls for improved national security safeguards.
“The AI public safety threat is no longer theoretical,” stated Sen. Hassan. “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.”