November 21, 2026 USD 44.93EUR 49.28PLN 11.06
Новини Чернігів — коротко й по суті
LIVE
Over half of damaged healthcare facilities restored in Chernihiv region Water damage in smartphones: 4 mistakes that can destroy your gadget EU Allocates €1 Billion for Military Equipment for Ukraine EU payments to Russia for energy exceed aid provided to Ukraine Russian forces strike kindergarten in Zaporizhzhia, causing fire in residential building Ukraine blocks Russian attempts to legitimize occupation at COP31 climate summit Missile strike on Kryvyi Rih: children and medic killed, dozens injured Zelenskyy: Crimean Bridge remains a legitimate target for the Armed Forces of Ukraine Russian attack on Zaporizhzhia: kindergarten and residential building damaged Energy infrastructure in Rivne region damaged: fifty settlements left without power Over half of damaged healthcare facilities restored in Chernihiv region Water damage in smartphones: 4 mistakes that can destroy your gadget EU Allocates €1 Billion for Military Equipment for Ukraine EU payments to Russia for energy exceed aid provided to Ukraine Russian forces strike kindergarten in Zaporizhzhia, causing fire in residential building Ukraine blocks Russian attempts to legitimize occupation at COP31 climate summit Missile strike on Kryvyi Rih: children and medic killed, dozens injured Zelenskyy: Crimean Bridge remains a legitimate target for the Armed Forces of Ukraine Russian attack on Zaporizhzhia: kindergarten and residential building damaged Energy infrastructure in Rivne region damaged: fifty settlements left without power

Study reveals vulnerabilities in popular chatbots: AI capable of generating attack plans

💡 Tech 1 min read 👁 0

US auditors have found that leading artificial intelligence models are capable of creating detailed terrorist attack scenarios following user manipulation.

Study reveals vulnerabilities in popular chatbots: AI capable of generating attack plans💡Tech

Photo: RBC-Ukraine

Key points

  • Chatbots bypassed their own safety restrictions in under 12 hours
  • Attack scenarios involved energy infrastructure and public events
  • US Senate plans investigation into AI developers

A large-scale study conducted by the US Government Accountability Office (GAO) between late 2025 and October 2026 has revealed significant security gaps in popular chatbots including ChatGPT, Claude, Gemini, and Grok. Experts proved that even users without specialized technical knowledge can bypass built-in system restrictions by gradually steering the conversation.

Manipulation mechanisms and test results

During the experiment, researchers simulated the behavior of average users. While the AI initially refused direct requests for attack plans, iterative questioning allowed the researchers to obtain instructions in under 12 hours. The report notes that the generated scenarios included specific advice on locations, timing, and methods for attacks, including the use of improvised explosive devices and drones to damage energy infrastructure.

Independent experts assessed these materials as realistic and capable of causing actual harm.

Senator Maggie Hassan has already announced congressional hearings regarding these risks. Technology developers, for their part, stated that they are continuing to improve safety mechanisms to prevent the use of AI for criminal purposes.

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.