November 21, 2026 USD 44.93EUR 49.28PLN 11.06
Новини Чорноморськ — коротко й по суті
LIVE
Healthcare Infrastructure Restoration in Chernihiv Region: Over Half of Facilities Repaired What to do if your smartphone gets wet: 4 common mistakes to avoid EU to provide €1 billion in military aid to Ukraine: key details EU Financial Dilemma: Payments to Russia for Energy Exceed Aid to Ukraine Another attack on Zaporizhzhia: kindergarten and residential building damaged Ukraine blocks Russia's attempts to legitimize occupied territories at COP31 climate summit Missile strike on Kryvyi Rih: civilians and a medic killed Zelenskyy Emphasizes Vulnerability of Crimean Bridge as Strategic Target Nighttime attack on Zaporizhzhia: kindergarten and residential building damaged Massive attack on the energy system: Rivne region suffers from power outages Healthcare Infrastructure Restoration in Chernihiv Region: Over Half of Facilities Repaired What to do if your smartphone gets wet: 4 common mistakes to avoid EU to provide €1 billion in military aid to Ukraine: key details EU Financial Dilemma: Payments to Russia for Energy Exceed Aid to Ukraine Another attack on Zaporizhzhia: kindergarten and residential building damaged Ukraine blocks Russia's attempts to legitimize occupied territories at COP31 climate summit Missile strike on Kryvyi Rih: civilians and a medic killed Zelenskyy Emphasizes Vulnerability of Crimean Bridge as Strategic Target Nighttime attack on Zaporizhzhia: kindergarten and residential building damaged Massive attack on the energy system: Rivne region suffers from power outages

Study reveals vulnerabilities in popular AI models: chatbots generated attack plans

💡 Tech 1 min read 👁 0

US auditors found that leading chatbots are capable of providing instructions for dangerous acts after being manipulated by users.

Study reveals vulnerabilities in popular AI models: chatbots generated attack plans💡Tech

Photo: RBC-Ukraine

Key points

  • AI models bypassed their own safety filters
  • Attack scenarios were generated in under 12 hours
  • US Congress plans hearings on AI safety

Risks of using generative AI

A recent report by the US Government Accountability Office (GAO) has uncovered significant security gaps in four leading chatbots, including ChatGPT, Claude, Gemini, and Grok. During a year-long testing process, researchers were able to obtain detailed attack plans despite the built-in safety filters.

The method to bypass these restrictions proved straightforward: instead of direct prompts, researchers used a series of follow-up questions. In less than 12 hours of interaction, the artificial intelligence generated scenarios including specific locations, timing, and methods of attack, such as the use of explosive devices or damage to energy infrastructure.

Response to audit results

Independent experts confirmed that the materials generated by the AI were realistic and posed a direct security threat. Senator Maggie Hassan has already announced congressional hearings, emphasizing the need for urgent action regarding these risks. Technology developers, in turn, stated they are continuing to work on strengthening the safety mechanisms of their platforms.

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.