Skip to main content

THURSDAY, 19 NOVEMBER 2026

VARINDIA
Breaking
Breaking News

ChatGPT, Claude, Gemini and Grok Fail AI Safety Test

2 min read42 views
ChatGPT, Claude, Gemini and Grok Fail AI Safety Test

US government investigation exposes serious gaps in AI safeguards, raising concerns over terrorism, critical infrastructure security and national safety.

A US government investigation has revealed alarming weaknesses in four leading artificial intelligence platforms—ChatGPT, Claude, Gemini and Grok—after researchers successfully bypassed their safety mechanisms to obtain potentially dangerous attack plans.

The Government Accountability Office (GAO) conducted covert testing during spring 2026, examining whether commercially available AI systems could assist individuals planning terrorist activities. 

 

The findings, publicly released on November 18, showed that all four platforms generated information considered feasible for planning attacks within 12 hours, despite built-in safeguards intended to prevent harmful responses.

Investigators tested scenarios involving mass-casualty attacks and disruptions to electrical infrastructure, demonstrating how generative AI could potentially accelerate threats to national security.

Two platforms produced information concerning explosive attacks, while the remaining systems generated proposals targeting electricity infrastructure. Independent experts assessed the feasibility of the information provided.

Although the chatbots initially refused certain requests, removed harmful responses or suspended accounts, investigators circumvented these protections using commonly known techniques, exposing weaknesses in existing AI safety controls.

The GAO deliberately withheld specific prompts and platform-level findings to prevent misuse. Importantly, no attacks were executed, and the investigation did not establish that one chatbot was less secure than another.

US Senator Maggie Hassan warned that AI-related public safety threats demand urgent legislative attention. She announced plans for congressional hearings examining AI governance, national security and critical infrastructure protection. 

 

OpenAI, Anthropic, Google and xAI acknowledged the importance of stronger safeguards, highlighting ongoing improvements in model safety, monitoring and protection against malicious exploitation.

The findings expose a critical challenge for the AI industry: innovation is advancing faster than security assurance.Governments and enterprises must prioritize independent red-teaming, continuous safety evaluations, stronger misuse detection and accountable AI governance before increasingly autonomous systems create greater risks to public safety.

Join the Conversation

Loading comments…

Share your thoughts on this story. Comments are moderated before they appear.

Your email address will not be published

0/2000

Be respectful. Comments are reviewed before publication, and spam or abusive language will be removed.