November 21, 2026 USD 44.93EUR 49.28PLN 11.06
Новини Ізюм — коротко й по суті
LIVE
Zelenskyy responds to Putin's claims regarding the Crimean Bridge's military use Zelenskyy and UK Defence Secretary Discuss Air Defense Enhancement and Long-Range Capabilities British medics arrange heartwarming farmyard reunion for critically ill farmer Denmark Seeks EU Visa Ban for Russians Following Thwarted Assassination Plot Against Ukrainian Rapper Denmark Urges EU to Tighten Schengen Entry Restrictions for Russian Citizens Amid Security Threats Volodymyr Zelenskyy Names Six Candidates for Ukrainian Ambassador to the United States Oleksii Honcharenko Criticizes Semen Kryvonos Over NABU Scandal President Zelensky comments on Ukraine's NATO membership prospects Zelenskyy Assesses Outcomes of Trilateral Summit in Shenzhen Ivan Kurlko Re-Elected Unanimously as President of the Ukrainian Biathlon Federation Zelenskyy responds to Putin's claims regarding the Crimean Bridge's military use Zelenskyy and UK Defence Secretary Discuss Air Defense Enhancement and Long-Range Capabilities British medics arrange heartwarming farmyard reunion for critically ill farmer Denmark Seeks EU Visa Ban for Russians Following Thwarted Assassination Plot Against Ukrainian Rapper Denmark Urges EU to Tighten Schengen Entry Restrictions for Russian Citizens Amid Security Threats Volodymyr Zelenskyy Names Six Candidates for Ukrainian Ambassador to the United States Oleksii Honcharenko Criticizes Semen Kryvonos Over NABU Scandal President Zelensky comments on Ukraine's NATO membership prospects Zelenskyy Assesses Outcomes of Trilateral Summit in Shenzhen Ivan Kurlko Re-Elected Unanimously as President of the Ukrainian Biathlon Federation

GAO report: popular AI chatbots vulnerable to generating dangerous attack scenarios

💡 Tech 1 min read 👁 0

US auditors found that leading AI models can be manipulated to plan attacks after bypassing their built-in safety mechanisms.

GAO report: popular AI chatbots vulnerable to generating dangerous attack scenarios💡Tech

Photo: RBC-Ukraine

Key points

  • GAO identified vulnerabilities in four major AI models
  • Attack scenarios were obtained in less than 12 hours
  • Developers are working to improve safety protocols
  • US Senate to hold hearings on AI risks

A comprehensive study by the US Government Accountability Office (GAO) has exposed significant security flaws in leading artificial intelligence systems. Over a nearly year-long testing period, researchers discovered that chatbots like ChatGPT, Claude, Gemini, and Grok could generate detailed attack plans if users employed a strategy of iterative, contextual prompting.

Bypassing safety protocols

While initial direct requests for dangerous instructions were blocked by safety filters, investigators managed to circumvent these restrictions in under 12 hours. By asking follow-up questions and guiding the conversation, they obtained specific scenarios, including instructions for creating explosive devices and methods for targeting energy infrastructure.

Independent experts assessed the generated materials as realistic and capable of causing genuine harm. Senator Maggie Hassan has announced upcoming Senate hearings, emphasizing the urgent need for regulation of technologies that pose national security risks. In response, AI developers stated they are continuing to enhance safety mechanisms to prevent such vulnerabilities in the future.

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.