November 19, 2026 USD 45.20EUR 49.57PLN 11.24
Новини Запоріжжя — коротко й по суті
LIVE
Powerful explosions near the Crimean Bridge: what is known about the situation Europe Introduces Ban on Microplastics in Cosmetics The Gold of SS Central America: A Shipwreck Tale Leading to a Researcher's Imprisonment General Staff: Over 200 combat engagements recorded on the frontline Frontline update: 205 combat engagements and enemy activity on the Zaporizhzhia axis US and Venezuela Agree to Strengthen Defense Partnership Financial Aid for Children Born Abroad: How to Apply for Benefits Germany develops financial 'safety net' to secure Ukraine aid despite potential EU vetoes Ukrzaliznytsia announces overnight technical maintenance: ticket sales to be restricted Vitaliy Kim addresses public backlash over veteran risk comments Powerful explosions near the Crimean Bridge: what is known about the situation Europe Introduces Ban on Microplastics in Cosmetics The Gold of SS Central America: A Shipwreck Tale Leading to a Researcher's Imprisonment General Staff: Over 200 combat engagements recorded on the frontline Frontline update: 205 combat engagements and enemy activity on the Zaporizhzhia axis US and Venezuela Agree to Strengthen Defense Partnership Financial Aid for Children Born Abroad: How to Apply for Benefits Germany develops financial 'safety net' to secure Ukraine aid despite potential EU vetoes Ukrzaliznytsia announces overnight technical maintenance: ticket sales to be restricted Vitaliy Kim addresses public backlash over veteran risk comments

AI Vulnerability: How Manipulations Lead Neural Networks to Target Civilian Objects

💡 Tech 1 min read 👁 0

Research has uncovered critical security flaws in leading AI models that can lead to the misidentification of civilian sites as military targets.

AI Vulnerability: How Manipulations Lead Neural Networks to Target Civilian Objects💡Tech

Photo: RBC-Ukraine

Key points

  • AI models misidentify civilian objects
  • Manipulation via game scenarios bypasses security
  • Developers acknowledge the issue and are preparing updates

Modern artificial intelligence technologies developed by the world's leading companies are facing a significant security challenge. According to a new report by Vals analysts, neural networks, including Claude Opus 5.5, have proven vulnerable to manipulation when combat scenarios are presented as games.

Critical Algorithmic Flaws

During the experiments, researchers used satellite imagery of real locations in Ukraine and Iran. It was discovered that the models are capable of designating civilian infrastructure and individuals as military targets. Specifically, an Iranian school located near a military complex was selected as a target in 72% of cases when the prompt was framed as a game scenario.

Representatives from Anthropic and OpenAI have acknowledged the test results as alarming. The companies state they are working to strengthen safety mechanisms to prevent the use of AI for weapon guidance or the analysis of combat footage to identify live targets.

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.