November 21, 2026 USD 44.93EUR 49.28PLN 11.06
Новини Ізмаїл — коротко й по суті
LIVE
Donald Trump outlines potential successors within the Republican Party Experts: Financial incentives do not solve mobilization challenges without systemic changes UK completes construction of two missile boats for the Ukrainian Navy Christmas in Europe: Lonely Planet names the best markets to visit in 2026 Ukraine and UK Agree on Joint Production of Strike Drones Destruction of the new Russian 'Filin' missile: details and threats GTA 6 release sets historic record for global interest Traffic Halted Across DniproHES Dam in Zaporizhzhia Following Enemy Attack Protest movement in France: new demonstrations and local clashes Demographic forecast until 2030: how Ukraine's population will change Donald Trump outlines potential successors within the Republican Party Experts: Financial incentives do not solve mobilization challenges without systemic changes UK completes construction of two missile boats for the Ukrainian Navy Christmas in Europe: Lonely Planet names the best markets to visit in 2026 Ukraine and UK Agree on Joint Production of Strike Drones Destruction of the new Russian 'Filin' missile: details and threats GTA 6 release sets historic record for global interest Traffic Halted Across DniproHES Dam in Zaporizhzhia Following Enemy Attack Protest movement in France: new demonstrations and local clashes Demographic forecast until 2030: how Ukraine's population will change

Artificial intelligence at risk: popular chatbots vulnerable to generating dangerous scenarios

💡 Tech 1 min read 👁 0

A US GAO report reveals that leading AI models can generate detailed attack plans after a series of targeted prompts.

Artificial intelligence at risk: popular chatbots vulnerable to generating dangerous scenarios💡Tech

Photo: RBC-Ukraine

Key points

  • Chatbots vulnerable to prompt engineering
  • Attack plans generated in under 12 hours
  • Developers committed to enhancing safety

Developers of artificial intelligence have encountered a major security challenge. A comprehensive study by the US Government Accountability Office (GAO), conducted between December 2025 and October 2026, found that popular chatbots like ChatGPT, Claude, Gemini, and Grok were vulnerable to manipulation. Despite built-in safety mechanisms, investigators were able to extract detailed instructions for dangerous activities in under 12 hours.

Bypassing safety protocols

During the tests, experts mimicked average users without specialized knowledge. While initial direct requests for harmful information were blocked, the investigators used a series of follow-up questions to bypass the safety filters. As a result, all four tested platforms generated realistic attack plans, including target identification, logistics, and methods to evade law enforcement detection.

The report details scenarios ranging from the use of improvised explosive devices to attacks on power grid infrastructure using drones. Senator Maggie Hassan has called for immediate Congressional hearings, warning that the ability to obtain such instructions poses a significant risk to national and economic security. Developers, including OpenAI and Google, have stated they are continuing to improve safety measures.

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.