November 21, 2026 USD 44.93EUR 49.28PLN 11.06
Новини Бар — коротко й по суті
LIVE
Healthcare Infrastructure Recovery in Chernihiv Region: Progress and Challenges EU to Allocate €1 Billion to Strengthen Ukraine's Defense Capabilities EU Spent More on Russian Energy Than on Aid to Ukraine, Commission Data Shows Another enemy attack on Zaporizhzhia: kindergarten and residential building damaged Ukraine blocks Russia's attempt to legitimize occupation at COP31 climate summit Rocket strike on Kryvyi Rih: casualties and dozens injured Zelenskyy states the Crimean Bridge will not survive Night attack on Zaporizhzhia: kindergarten and residential building damaged Rivne region under attack: fifty settlements left without power How to stay warm at home this winter: expert tips for Bar residents Healthcare Infrastructure Recovery in Chernihiv Region: Progress and Challenges EU to Allocate €1 Billion to Strengthen Ukraine's Defense Capabilities EU Spent More on Russian Energy Than on Aid to Ukraine, Commission Data Shows Another enemy attack on Zaporizhzhia: kindergarten and residential building damaged Ukraine blocks Russia's attempt to legitimize occupation at COP31 climate summit Rocket strike on Kryvyi Rih: casualties and dozens injured Zelenskyy states the Crimean Bridge will not survive Night attack on Zaporizhzhia: kindergarten and residential building damaged Rivne region under attack: fifty settlements left without power How to stay warm at home this winter: expert tips for Bar residents

GAO report: popular chatbots vulnerable to generating attack plans

💡 Tech 1 min read 👁 1

US auditors found that leading artificial intelligence models can generate detailed attack plans after being manipulated by users.

GAO report: popular chatbots vulnerable to generating attack plans💡Tech

Photo: RBC-Ukraine

Key points

  • GAO identified vulnerabilities in four leading AI models
  • Follow-up questioning techniques can bypass safety restrictions
  • Developers are working to enhance system security

Security risks in the AI sector

According to a report by the US Government Accountability Office (GAO), popular chatbots, including ChatGPT, Claude, Gemini, and Grok, have proven vulnerable to prompts regarding the planning of illegal activities. During a comprehensive testing period that lasted nearly a year, researchers were able to bypass built-in system safeguards in less than 12 hours of interaction.

The experiment demonstrated that direct requests for dangerous content are typically blocked by algorithms. However, by employing a method of follow-up questions and gradually steering the AI toward the topic, investigators were able to obtain detailed instructions. These included plans for attacking energy infrastructure, scenarios for using explosive devices, and advice on evading law enforcement.

Response to identified threats

Independent experts confirmed that the materials generated by the AI were realistic and potentially harmful. Senator Maggie Hassan has already announced upcoming congressional hearings, emphasizing the urgent need for stricter oversight of technology developers. The companies involved have stated that they continue to work on improving their safety mechanisms.

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.