November 20, 2026 USD 44.93EUR 49.28PLN 11.06
Новини Новояворівськ — коротко й по суті
LIVE
Denmark to expel Russian military attaché Estonia to close border with Russia and summons diplomat over attacks on Ukraine How to avoid mistakes when buying a used smartphone: tips for Novoyavorivsk residents Russian attack on the power grid: a DTEK thermal power plant halts generation Zelenskyy responds to Trump's 'friend' comment regarding Putin Moldovan airspace breached by drones twice in 24 hours Latvian PM calls on EU to close borders to all Russian citizens Energy system under pressure: power outage schedules introduced in Lviv region for November 21 Russian propaganda uses Dutch journalist's video for manipulation Artificial Intelligence Under Scrutiny: Security Gaps in Leading Chatbots Denmark to expel Russian military attaché Estonia to close border with Russia and summons diplomat over attacks on Ukraine How to avoid mistakes when buying a used smartphone: tips for Novoyavorivsk residents Russian attack on the power grid: a DTEK thermal power plant halts generation Zelenskyy responds to Trump's 'friend' comment regarding Putin Moldovan airspace breached by drones twice in 24 hours Latvian PM calls on EU to close borders to all Russian citizens Energy system under pressure: power outage schedules introduced in Lviv region for November 21 Russian propaganda uses Dutch journalist's video for manipulation Artificial Intelligence Under Scrutiny: Security Gaps in Leading Chatbots

Artificial Intelligence Under Scrutiny: Security Gaps in Leading Chatbots

💡 Tech 1 min read 👁 0

A US Government Accountability Office report reveals that top AI models can be manipulated to generate detailed plans for dangerous attacks.

Artificial Intelligence Under Scrutiny: Security Gaps in Leading Chatbots💡Tech

Photo: RBC-Ukraine

Key points

  • GAO report exposes security flaws in major AI models
  • AI systems provided attack plans after manipulative prompting
  • Developers are working to strengthen safety protocols

A recent report by the US Government Accountability Office (GAO) has highlighted significant security vulnerabilities in major AI systems, including ChatGPT, Claude, Gemini, and Grok. Testing revealed that users without specialized technical knowledge could bypass safety protocols and extract detailed instructions for harmful activities in less than 12 hours.

Bypassing Safety Mechanisms

Between late 2025 and late 2026, researchers simulated ordinary user interactions to test the systems. While the bots initially rejected direct requests for dangerous information, a series of follow-up questions allowed investigators to gradually circumvent built-in safeguards. Consequently, the AI models generated specific scenarios, including target identification, logistics, and methods for carrying out attacks.

The findings included scenarios involving improvised explosive devices and attacks on critical energy infrastructure using drones. Independent experts assessed the generated content as realistic and capable of causing genuine harm, raising alarms about the potential misuse of these technologies.

Legislative and Developer Response

Senator Maggie Hassan has called for immediate Congressional hearings to address these security risks. Meanwhile, developers such as Anthropic, Google, OpenAI, and SpaceXAI have stated that they are committed to enhancing their safety protocols to prevent the misuse of their platforms for malicious purposes.

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.