November 21, 2026 USD 44.93EUR 49.28PLN 11.06
Новини Павлоград — коротко й по суті
LIVE
Head of state considers six candidates for the position of Ukraine's ambassador to the United States Goncharenko accuses NABU Director Kryvonos of undermining the agency's reputation The Revolution of Dignity and Media Breakthrough: How a Social Media Post Changed History Zelenskyy on Shenzhen Summit: Putin's Attempt to Redraw Borders Fails Ivan Krulko unanimously re-elected as President of the Ukrainian Biathlon Federation Russian cargo ship on fire in the Gulf of Finland after drone attack Volodymyr Zelenskyy Discusses Prospects of a Trilateral Summit at the G20 Zelenskyy Emphasizes Need for Legal Security Guarantees for Ukraine Strike on Kerch Bridge: European Intelligence Evaluates Impact on Russian Logistics Winter Energy Efficiency: German Ventilation Method for Pavlohrad Residents Head of state considers six candidates for the position of Ukraine's ambassador to the United States Goncharenko accuses NABU Director Kryvonos of undermining the agency's reputation The Revolution of Dignity and Media Breakthrough: How a Social Media Post Changed History Zelenskyy on Shenzhen Summit: Putin's Attempt to Redraw Borders Fails Ivan Krulko unanimously re-elected as President of the Ukrainian Biathlon Federation Russian cargo ship on fire in the Gulf of Finland after drone attack Volodymyr Zelenskyy Discusses Prospects of a Trilateral Summit at the G20 Zelenskyy Emphasizes Need for Legal Security Guarantees for Ukraine Strike on Kerch Bridge: European Intelligence Evaluates Impact on Russian Logistics Winter Energy Efficiency: German Ventilation Method for Pavlohrad Residents

Study reveals vulnerabilities in AI chatbots: systems generated attack plans

💡 Tech 1 min read 👁 1

Extensive testing has shown that leading AI models can be manipulated to provide dangerous instructions within hours.

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

Photo: RBC-Ukraine

Key points

  • Top AI models failed to block requests for attack planning.
  • Safety filters were bypassed in under 12 hours.
  • U.S. Senate to hold hearings on AI safety risks.

Security gaps in generative AI

A recent report by the U.S. Government Accountability Office (GAO) has highlighted significant security flaws in four major AI chatbots: ChatGPT, Claude, Gemini, and Grok. Over a period of testing, researchers posing as average users successfully bypassed safety filters to extract harmful information.

The findings indicate that within 12 hours of interaction, the systems were able to generate detailed attack scenarios. These included instructions for creating explosive devices and targeting critical infrastructure, such as power grids. Experts have classified the generated material as realistic and potentially dangerous.

Calls for stricter regulation

Senator Maggie Hassan has announced upcoming Senate hearings to address these risks, urging Congress to act swiftly. Developers including Anthropic, Google, OpenAI, and SpaceXAI have stated they are continuing to enhance safety protocols. However, the incident has sparked a global debate regarding the balance between innovation and public safety.

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.