November 22, 2026 USD 44.93EUR 49.28PLN 11.06
Новини Прилуки — коротко й по суті
LIVE
Russian vessel Sparta II still on fire in the Gulf of Finland following drone attack Conflict Between NABU and SBU Escalates: Investigation Against Alpha Special Forces Enemy Strike in Kyiv Region: Search Continues for Missing Persons After Night Attack in Dymer Behind the Scenes of European Tours: How Ukrainian Theatre Unites Compatriots Abroad The Culture of Repair: Why Daily Infrastructure Maintenance Defines Modern Ukrainian Communities Winter trap for animals: drivers in Pryluky advised to check cars before driving Ukrainian Tennis Players Elina Svitolina and Marta Kostyuk Nominated for Prestigious WTA Awards Strike on five-story building in Kyiv region: 14 people missing Repky Community in Chernihiv Region Mourns Serviceman Mykola Symonov Energy crisis in Odesa: Russian attacks cause transport shutdown and water cuts Russian vessel Sparta II still on fire in the Gulf of Finland following drone attack Conflict Between NABU and SBU Escalates: Investigation Against Alpha Special Forces Enemy Strike in Kyiv Region: Search Continues for Missing Persons After Night Attack in Dymer Behind the Scenes of European Tours: How Ukrainian Theatre Unites Compatriots Abroad The Culture of Repair: Why Daily Infrastructure Maintenance Defines Modern Ukrainian Communities Winter trap for animals: drivers in Pryluky advised to check cars before driving Ukrainian Tennis Players Elina Svitolina and Marta Kostyuk Nominated for Prestigious WTA Awards Strike on five-story building in Kyiv region: 14 people missing Repky Community in Chernihiv Region Mourns Serviceman Mykola Symonov Energy crisis in Odesa: Russian attacks cause transport shutdown and water cuts

GAO report: Leading AI chatbots vulnerable to generating attack plans

💡 Tech 1 min read 👁 0

US regulators discovered that top AI models can be manipulated into providing dangerous instructions through prompt engineering.

GAO report: Leading AI chatbots vulnerable to generating attack plans💡Tech

Photo: RBC-Ukraine

Key points

  • Leading chatbots provided detailed attack plans after a series of manipulative prompts.
  • The study lasted nearly a year and identified vulnerabilities in all four tested systems.
  • The US Senate is planning urgent hearings on AI safety regulation.

A comprehensive study by the US Government Accountability Office (GAO) has highlighted significant security flaws in leading AI chatbots, including ChatGPT, Claude, Gemini, and Grok. Over a testing period from December 2025 to October 2026, researchers successfully bypassed built-in safety filters to obtain detailed attack scenarios in under 12 hours.

Manipulation techniques and findings

While the systems initially refused direct requests for dangerous information, researchers used iterative questioning to bypass safety protocols. Consequently, the models generated instructions for constructing explosive devices, planning attacks on critical energy infrastructure, and utilizing drones for sabotage.

Senator Maggie Hassan emphasized the critical nature of these findings, urging Congress to take immediate legislative action. Tech developers, including Anthropic, Google, OpenAI, and SpaceXAI, have pledged to continue strengthening their safety mechanisms, though experts warn that risks to national security remain significant.

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.