Federal investigators working for the Government Accountability Office (GAO) reported that they were able to bypass safety safeguards on four major artificial‑intelligence platforms and obtain detailed plans for potentially deadly terrorist attacks in less than 12 hours.
Testing revealed alarming capabilities
The undercover testing, conducted in the spring of 2026, targeted ChatGPT, Claude, Gemini and Grok. Each system generated information that could help a malicious actor carry out attacks on crowded public venues or on the nation’s electrical grid.
Two of the platforms produced step‑by‑step instructions for constructing improvised explosive devices intended to cause mass casualties. The other two outlined strategies for disrupting electrical infrastructure, aiming to inflict significant economic damage.
All four AI responses included specific potential targets, locations, timing and methods of attack. Although each platform initially activated safeguards designed to block or remove harmful content, GAO investigators were ultimately able to circumvent those protections.
Feasibility confirmed by experts
GAO investigators consulted independent experts and performed real‑world checks on the generated plans. They concluded that the information provided by each AI system was sufficiently feasible to assist someone planning an attack.
“The potential benefits of this technology are immense, but they can also benefit malicious actors,” the GAO warned, emphasizing growing risks to national and economic security.
Company responses
The companies behind the platforms—OpenAI, Anthropic, Google and SpaceXAI—acknowledged the findings and said they are continuing to improve safety measures.
Anthropic noted that determined users could bypass safeguards and said it took the findings seriously, while also questioning aspects of the GAO’s methodology and stating it could not independently verify the results.
Google said investigators had tested earlier versions of its models and added, “We continue to improve our models.” OpenAI welcomed the opportunity to respond and recognized “the work that remains,” and SpaceXAI said it “continues to work diligently” to strengthen safety with each model release.
Legislative interest
The report was requested by Senator Maggie Hassan (D‑NH), who highlighted the urgency of the issue. “The AI public safety threat is no longer theoretical,” Hassan said in a statement. “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.”
The GAO’s findings add to growing scrutiny of artificial‑intelligence misuse and renewed calls for stronger federal safeguards against national‑security threats.
Original reporting: Fox News (HLL/CB) — read the source article.