November 18, 2026 USD 45.20EUR 49.57PLN 11.24
Новини Чернігів — коротко й по суті
LIVE
Study reveals AI model vulnerability to manipulation in military scenarios Energy forecast: what Chernihiv residents should expect in the coming months Ukraine national football team to learn Euro 2028 qualifying opponents: draw details Sports Isolation Challenged: How Nations Respond to the Return of Russian Athletes Chernihiv artisan creates unique handmade herbal dolls filled with local plants Diplomatic protocol at the summit in China: details of the leaders' photo Official Exchange Rates: NBU Strengthens Hryvnia Against Dollar and Euro Berlin hosts 'Coalition of the Willing' to discuss multinational forces in Ukraine Chernihiv Families Encouraged to Prepare Letters to Saint Nicholas in Advance OLX to Introduce Mandatory Verification via 'Diia': What Users Should Know Study reveals AI model vulnerability to manipulation in military scenarios Energy forecast: what Chernihiv residents should expect in the coming months Ukraine national football team to learn Euro 2028 qualifying opponents: draw details Sports Isolation Challenged: How Nations Respond to the Return of Russian Athletes Chernihiv artisan creates unique handmade herbal dolls filled with local plants Diplomatic protocol at the summit in China: details of the leaders' photo Official Exchange Rates: NBU Strengthens Hryvnia Against Dollar and Euro Berlin hosts 'Coalition of the Willing' to discuss multinational forces in Ukraine Chernihiv Families Encouraged to Prepare Letters to Saint Nicholas in Advance OLX to Introduce Mandatory Verification via 'Diia': What Users Should Know

Study reveals AI model vulnerability to manipulation in military scenarios

💡 Tech 2 min read 👁 0

Analysts found that modern neural networks can identify civilian objects as military targets if the prompt is framed as a game.

Study reveals AI model vulnerability to manipulation in military scenarios💡Tech

Photo: RBC-Ukraine

Key points

  • AI models mistakenly mark civilian objects as targets
  • Manipulation via 'game' framing bypasses safety filters
  • Developers acknowledge the issue and are strengthening safeguards

Risks of AI in military applications

A new study by Vals has demonstrated critical security gaps in leading artificial intelligence models, such as Claude Opus 5.5, OpenAI, and Google. Experiments revealed that neural networks are capable of misclassifying civilian objects, including schools and people, as potential military targets.

Researchers utilized real satellite imagery and drone footage to test the AI's ability to perform autonomous analysis. It turned out that the models' safety filters are easily bypassed: one only needs to add a statement to the prompt that the events are part of a "game." Under these conditions, the Claude Opus 5.5 model selected a civilian school as a target in 72% of test cases.

Developer response and security threats

Representatives from the developing companies, including Anthropic and OpenAI, acknowledged the findings as unacceptable and announced work on strengthening safety algorithms. Despite technical limitations, such as a lack of onboard computing power for drones, experts warn of the danger of using publicly available AI tools to process large amounts of battlefield data.

There is a risk that data obtained through manipulation could be used to train automated weapon-targeting systems. This highlights the need for strict control over the development of technologies capable of influencing decision-making in critical situations.

Source RBC-Ukraine — rbc.ua/ukr/news/dostatno-nazvati-boyovi-diyi-groyu-claude-179501202… ↗

Contact the newsroom

Got a tip, a proposal or a partnership question? Write to us and we will reply shortly.