Google and NATO Warn of Rising AI Cyber Threats, Including Model Theft and Accelerated Attacks

Key Intelligence

• ■Google has issued warnings that state-backed groups and other malicious actors are actively misusing AI models, including Gemini, across various stages of cyberattacks to enhance speed and sophistication.
• ■Threat actors are attempting to steal proprietary AI models, evidenced by hackers "hammering" Google's Gemini with thousands of prompts to clone its large language model, and accusations against companies like DeepSeek for distilling US models.
• ■The emergence of open-source AI agents, such as OpenClaw, is identified as expanding the attack surface and introducing new security risks.
• ■NATO has cautioned against AI model poisoning as a tactic in information warfare, highlighting a broader spectrum of AI-related threats beyond direct cyber exploitation.
• ■Critical AI vulnerabilities are reportedly being exploited at a rate that outpaces current defensive capabilities, raising concerns about the rapid evolution of AI-powered threats.