GAO tests got Claude, ChatGPT, Gemini and Grok to produce viable attack plans
Covert tests by the US Government Accountability Office got the four most used generative AI platforms, Anthropic's Claude, SpaceX AI's Grok, OpenAI's ChatGPT and Google's Gemini, to produce feasible information for planning and carrying out attacks, according to a report the congressional watchdog released on Wednesday.
In tests run in spring 2026, two platforms gave information for building and detonating a vehicle bomb designed to cause mass casualties, naming a specific hotel and a specific convention center in major US cities, with exact dates and times. The other two produced plans to attack electrical grid substations with a drone and with firearms. Each platform took less than 12 hours to produce its plan.
The prompts triggered each platform's safeguards, including refusals, removed answers and suspended accounts, but GAO got around them with common bypass techniques. It checked the plans through research, surveillance and interviews with experts, including a bomb squad commander, ABC News reports. GAO did not say which platform produced which result.
The companies told GAO they continually test their systems, add safeguards and have improved their models since the tests. "However, we conclude that, in its current state, this technology can be dangerous, and its misuse represents a growing risk to our nation," said Kevin Walsh, GAO's director of IT and cybersecurity audits.
Sen. Maggie Hassan, who requested the study and is in line to chair the Senate Homeland Security and Governmental Affairs Committee, said she will seek stricter guardrails and hold at least one AI safety hearing early next year. "If Congress doesn't act, it is a question of when, not if, a bad actor will use AI to plan and carry out a major attack," she said, according to MS NOW.
We hope you enjoyed this article
Consider subscribing to one of our newsletters like AI Policy Brief, Cybersecurity AI Weekly or Daily AI Brief.
Also, consider following us on social media:
More from Regulation
Nov 18 Singapore sets up sandbox for AI agent payments under new national AI missions for finance Nov 18 China's military is seeking Nvidia's most advanced chips, US congressional commission says Nov 18 North Carolina becomes the tenth state to offer Google's free AI training Nov 18 LeBron James sues AI betting app Bron over ad using his likeness and voice Nov 18 San Joaquin County extends its AI data centre moratorium by nearly a yearMore from Cybersecurity
Nov 18 Canonical lets Ubuntu Pro users install AI-written security patches before human review Nov 18 Malicious versions of Microsoft's agent-framework package on PyPI stole Azure and CI credentials, Wiz says Nov 18 Taylor & Francis investigates journal paper with prompts telling AI reviewers to write a positive review Nov 18 Reflection AI limits its new Laser hacking tool to vetted organisations Nov 18 Barracuda says attackers used ARTEX AI framework and a stolen vendor key to breach its cloudCybersecurity AI Weekly
Weekly newsletter about AI in Cybersecurity.
Whitepaper
Tensordyne Napier: What If One Rack Could Do the Work of Nine?
Tensordyne
This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.
Read moreYou may also like
Geoffrey Hinton Warns of Major AI Incident by 2027
Nonprofits sue Anthropic, Google, Meta, OpenAI and SpaceX to halt development of new AI models
Senator Mark Kelly proposes safety standards and independent examiners for frontier AI
OpenAI Urges California to Expand AI Incident Reporting
Cruz postpones classified Senate briefing on AI security incidents
Daily AI Brief: the AI news that matters, in your inbox.