November 21, 2026 USD 44.93EUR 49.28PLN 11.06
Новини Шептицький — коротко й по суті
LIVE
Healthcare Infrastructure Recovery in Chernihiv Region: Over Half of Damaged Facilities Restored What to do if your smartphone falls into water: 4 common mistakes to avoid EU to allocate 1 billion euros for military equipment for Ukraine EU spent more on Russian energy than on support for Ukraine, EC data shows Another attack on Zaporizhzhia: kindergarten and residential building damaged Ukraine blocks Russia's attempts to legitimize occupation at COP31 climate conference Missile strike on Kryvyi Rih: civilians and a medic killed Zelenskyy reaffirms legitimacy of strikes on Crimean Bridge Russian attack on Zaporizhzhia: kindergarten and residential building damaged Rivne region energy infrastructure damaged by massive Russian attack Healthcare Infrastructure Recovery in Chernihiv Region: Over Half of Damaged Facilities Restored What to do if your smartphone falls into water: 4 common mistakes to avoid EU to allocate 1 billion euros for military equipment for Ukraine EU spent more on Russian energy than on support for Ukraine, EC data shows Another attack on Zaporizhzhia: kindergarten and residential building damaged Ukraine blocks Russia's attempts to legitimize occupation at COP31 climate conference Missile strike on Kryvyi Rih: civilians and a medic killed Zelenskyy reaffirms legitimacy of strikes on Crimean Bridge Russian attack on Zaporizhzhia: kindergarten and residential building damaged Rivne region energy infrastructure damaged by massive Russian attack

GAO Study: Popular AI Chatbots Found Vulnerable to Generating Dangerous Scenarios

💡 Tech 2 min read 👁 0

U.S. auditors found that leading AI models can generate attack plans after a series of follow-up prompts.

GAO Study: Popular AI Chatbots Found Vulnerable to Generating Dangerous Scenarios💡Tech

Photo: RBC-Ukraine

Key points

  • GAO identified vulnerabilities in four leading chatbots
  • Systems generated attack plans in under 12 hours
  • U.S. Senate plans hearings on AI safety

Vulnerability of Artificial Intelligence Systems

A recent report by the U.S. Government Accountability Office (GAO) has identified critical flaws in the safety mechanisms of leading chatbots, including ChatGPT, Claude, Gemini, and Grok. Between December 2025 and October 2026, researchers tested whether these tools could provide instructions for harmful activities. The findings revealed that even users without specialized knowledge could bypass built-in restrictions through a series of sequential prompts.

Detailed Attack Scenarios

During the experiment, all four tested systems, labeled A through D in the report, generated detailed attack scenarios in less than 12 hours. These instructions included target selection, timing, methods for acquiring materials, and tactics to evade law enforcement. Specifically, some models offered plans for using explosive devices or attacking energy infrastructure, including the use of drones.

An attacker could use a standard AI chatbot and, in less than 12 hours, obtain a ready-made guide for carrying out a lethal attack, — stated Senator Maggie Hassan.

Developers, including OpenAI, Google, Anthropic, and SpaceXAI, stated they are continuing to improve their safety mechanisms. Meanwhile, the U.S. Congress plans to hold hearings on strengthening the regulation of artificial intelligence technologies to prevent threats to national security.

Source RBC-Ukraine — rbc.ua/ukr/news/chatgpt-claude-gemini-ta-grok-stvorili-plani-179518… ↗

Contact the newsroom

Got a tip, a proposal or a partnership question? Write to us and we will reply shortly.