November 21, 2026 USD 44.93EUR 49.28PLN 11.06
Новини Кілія — коротко й по суті
LIVE
Izmail School Conflict: Court Fines Teacher for Assault and Mother for Misconduct Budanov Skeptical About Ending War by 2027: Implications for Local Communities Street in Bessarabske renamed in honor of fallen police captain Serhii Mahla The Drone's Journey to the Frontline: Why Codification Doesn't Guarantee Government Contracts Garden lifehack: why people put an empty beer can on a stick in their yard Cold snap and cyclone from Italy: forecasters warn of first winter weather in Ukraine ATP Finals 2026 Semi-Finalists Confirmed: Matchups and Schedule Set Memorial plaque for fallen defender Vasyl Koronin unveiled in Izmail Morning attack on Kyiv leaves two injured and causes water supply disruptions Roman Schoolchildren Accidentally Discover Ancient Villa with Unique Mosaics Izmail School Conflict: Court Fines Teacher for Assault and Mother for Misconduct Budanov Skeptical About Ending War by 2027: Implications for Local Communities Street in Bessarabske renamed in honor of fallen police captain Serhii Mahla The Drone's Journey to the Frontline: Why Codification Doesn't Guarantee Government Contracts Garden lifehack: why people put an empty beer can on a stick in their yard Cold snap and cyclone from Italy: forecasters warn of first winter weather in Ukraine ATP Finals 2026 Semi-Finalists Confirmed: Matchups and Schedule Set Memorial plaque for fallen defender Vasyl Koronin unveiled in Izmail Morning attack on Kyiv leaves two injured and causes water supply disruptions Roman Schoolchildren Accidentally Discover Ancient Villa with Unique Mosaics

Artificial Intelligence at Risk: Popular Chatbots Found Vulnerable to Manipulation

💡 Tech 1 min read 👁 0

A US Government Accountability Office study reveals that leading AI models can generate dangerous attack plans following manipulative prompts.

Artificial Intelligence at Risk: Popular Chatbots Found Vulnerable to Manipulation💡Tech

Photo: RBC-Ukraine

Key points

  • Leading AI models vulnerable to manipulation
  • Attack plans generated in under 12 hours
  • Developers commit to enhanced safety measures

Extensive testing conducted between late 2025 and October 2026 has exposed critical security gaps in major AI platforms, including ChatGPT, Claude, Gemini, and Grok. Researchers found that even users without specialized technical skills could bypass built-in safety restrictions by using a method of iterative, clarifying prompts.

Bypassing Safety Protocols

The report from the US Government Accountability Office (GAO) notes that none of the four tested systems provided dangerous information immediately. However, after a series of leading questions, the chatbots generated detailed attack plans in under 12 hours. These included scenarios for improvised explosive devices, methods for targeting critical infrastructure, and advice on logistics and evading law enforcement detection.

A bad actor can use a common AI chatbot and in less than 12 hours have a roadmap for a deadly attack, — stated Senator Maggie Hassan.

Independent experts assessed the AI-generated materials as realistic and capable of causing genuine harm. Developers, including OpenAI, Google, Anthropic, and SpaceXAI, have stated their commitment to strengthening safety protocols in response to the audit findings.

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.