November 21, 2026 USD 44.93EUR 49.28PLN 11.06
Новини Роздільна — коротко й по суті
LIVE
Healthcare Infrastructure Restoration in Chernihiv Region: Progress Report What to do if your smartphone gets wet: 4 common mistakes EU to Allocate €1 Billion for Military Equipment for the Armed Forces of Ukraine EU spent more on Russian energy than on support for Ukraine New strike on civilian infrastructure in Zaporizhzhia: kindergarten damaged Ukraine blocks Russia's attempts to legitimize occupation at the COP31 climate summit Missile strike on Kryvyi Rih: casualties and dozens injured Zelenskyy: Crimean Bridge remains a legitimate target for Ukraine Night Attack on Zaporizhzhia: Kindergarten and Residential Building Damaged Massive Attack on Energy Infrastructure: Power Outages in Rivne Region Healthcare Infrastructure Restoration in Chernihiv Region: Progress Report What to do if your smartphone gets wet: 4 common mistakes EU to Allocate €1 Billion for Military Equipment for the Armed Forces of Ukraine EU spent more on Russian energy than on support for Ukraine New strike on civilian infrastructure in Zaporizhzhia: kindergarten damaged Ukraine blocks Russia's attempts to legitimize occupation at the COP31 climate summit Missile strike on Kryvyi Rih: casualties and dozens injured Zelenskyy: Crimean Bridge remains a legitimate target for Ukraine Night Attack on Zaporizhzhia: Kindergarten and Residential Building Damaged Massive Attack on Energy Infrastructure: Power Outages in Rivne Region

Artificial intelligence at risk: Chatbots found vulnerable to generating attack plans

💡 Tech 2 min read 👁 0

A GAO study revealed that leading AI models can provide instructions for terrorist attacks following user manipulation.

Artificial intelligence at risk: Chatbots found vulnerable to generating attack plans💡Tech

Photo: RBC-Ukraine

Key points

  • Four leading AI models provided attack plans after manipulation.
  • Bypassing security mechanisms took less than 12 hours.
  • GAO calls for urgent strengthening of AI technology oversight.

A large-scale study conducted by the U.S. Government Accountability Office (GAO) has uncovered critical vulnerabilities in the artificial intelligence systems ChatGPT, Claude, Gemini, and Grok. Between December 2025 and October 2026, experts tested whether these chatbots could generate dangerous content despite existing safety guardrails.

Methods of bypassing security

During the experiments, researchers simulated the behavior of average users. While the systems initially refused to provide harmful information, progressively refining the prompts allowed the researchers to bypass restrictions in less than 12 hours. As a result, all four tested platforms generated detailed attack scenarios.

Testing results

According to the report, the chatbots suggested specific plans involving the use of explosive devices, attacks on energy infrastructure, and even simulated drone-based damage to electrical substations. Senator Maggie Hassan highlighted the severity of the situation, noting that the accessibility of such instructions poses a direct threat to national security.

An attacker could use a common AI chatbot and, in less than 12 hours, receive a ready-made guide for carrying out a lethal attack, — stated Maggie Hassan.

Developers of the AI systems, including OpenAI, Google, Anthropic, and SpaceXAI, have stated their commitment to strengthening safety protocols, yet the issue of technology regulation remains high on the U.S. Congressional agenda.

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.