AI Vulnerabilities: Chatbots Found Capable of Generating Attack Plans
A US Government Accountability Office report reveals that leading AI models can be manipulated into providing dangerous instructions for attacks.
💡TechPhoto: RBC-Ukraine
Key points
- Bypassing AI safety guardrails
- Generation of realistic attack scenarios
- Urgent call for Congressional oversight
A recent report by the US Government Accountability Office (GAO) has highlighted significant security flaws in major AI platforms, including ChatGPT, Claude, Gemini, and Grok. Following extensive testing between late 2025 and 2026, researchers demonstrated that built-in safety guardrails can be bypassed through persistent, iterative questioning.
Security Concerns and Legislative Response
While the AI models initially refused direct requests for harmful information, investigators found that by using contextual follow-up questions, they could extract detailed attack scenarios. These included instructions for creating explosive devices and targeting critical energy infrastructure. The results were deemed realistic enough to pose a genuine threat.
Senator Maggie Hassan has called for immediate Congressional hearings to address these risks. Major AI developers have responded by stating that they are actively working to strengthen their safety protocols. The GAO report emphasizes that while AI offers immense potential, its misuse for illegal activities presents a growing challenge to national and economic security.
Source RBC-Ukraine — rbc.ua/ukr/news/chatgpt-claude-gemini-ta-grok-stvorili-plani-179518… ↗
Read next
EU to Allocate €1 Billion to Strengthen Ukraine's Defense Capabilities
The European Union has mobilized an additional €1 billion through the European Peace Facility to supply military equipment to the Ukrainian Armed Forces by the end of 202…
EU spent more on Russian energy than on aid to Ukraine, report finds
An analysis of EU expenditures shows that since 2022, payments for energy imports from Russia have exceeded the total financial support provided to Ukraine.
Color Preferences and Intelligence: Scientific Insights
Research indicates a connection between clothing color choices, personality types, and cognitive abilities.
Russian attack on Zaporizhzhia: kindergarten and residential building damaged
A Russian strike on Zaporizhzhia targeted a kindergarten, causing a fire in a nearby residential building.
Ukraine blocks Russian attempts to legitimize occupation of territories at COP31 climate summit
The Ukrainian delegation at the UN conference in Antalya opposed the inclusion of data from occupied regions in Russian reports.
🛡️SecurityMissile strike on Kryvyi Rih leaves civilians and medic dead
A ballistic missile attack on the center of Kryvyi Rih has caused significant casualties and damage to residential infrastructure.