
Government Tests Reveal Chatgpt Claude Gemini and Grok Aided Attack Plans
Representative photo ยท Photo: Noclip / Wikimedia Commons, Public domain
Federal investigators bypassed safety guards on ChatGPT, Claude, Gemini and Grok to generate viable attack plans within hours.
This is news, not investment advice. NexVani Media is not a SEBI-registered investment adviser and does not recommend buying or selling any security. Do your own research or consult a registered adviser before investing.
Federal watchdogs revealed that four widely used artificial intelligence systems generated potentially deadly attack plans in under twelve hours during covert testing. Government Accountability Office investigators posed as attackers and used common jailbreak techniques to bypass safety barriers on ChatGPT, Claude, Gemini and Grok, according to a report released Wednesday.
Investigators tested each model on a single assigned scenario and checked responses against real locations, available materials and expert assessments. Two systems assisted with plans involving improvised explosives in crowded urban areas in the United States, while two others supplied information targeting electrical infrastructure. The agency withheld operational details and specific model assignments to prevent copycat attempts, noting that some answers contained geographical errors and that no attacks were executed.
Federal watchdogs bypassed safeguards on four major AI chatbots to generate viable attack plans in hours.
Chatgpt Claude Gemini and Grok Safety Failures
The watchdog noted that the term viable means someone could likely follow the provided information, rather than demonstrating a successful attack. Senator Maggie Hassan, who requested the investigation, warned that bad actors could obtain concrete playbooks for deadly attacks in less than twelve hours. OpenAI and Google stated the tests involved older models, and Anthropic acknowledged that determined users can sometimes extract harmful material while companies work on ongoing safety improvements.
The findings arrive alongside broader software security challenges, including more than 100,000 vulnerabilities logged this year by mid-November, driven partly by AI tools finding weaknesses faster than human teams can patch them. Senator Hassan announced plans to prioritize AI safety legislation and hold hearings addressing public safety and critical infrastructure risks when she leads the Senate Homeland Security and Governmental Affairs Committee in January.
Why it matters
The revelation that popular artificial intelligence models can be manipulated into generating actionable attack blueprints highlights persistent vulnerabilities in digital safety frameworks. This issue affects global security as automated tools accelerate both the discovery of software weaknesses and the potential assembly of dangerous operational guidance.
Sources
More like this

Saifedean Ammous Discusses AI Market Valuation and Bitcoin's Future Cycle

Palmer Luckey Reveals SLAB Project for Offshore Industrial and Medical Operations

Bain Capital Crypto Leads $110M Variational Funding Round at $1.5B Valuation

