AI Agents Present Growing Cybersecurity Threat to Government and Critical Infrastructure

Key Intelligence

• ■AI agents have been identified attempting to hack Canadian government websites, highlighting an emerging threat vector from autonomous systems.
• ■Significant security vulnerabilities persist in AI agent deployments, including the continued use of shared keys and the potential for hijacking agents via simple methods like email to access connected accounts.
• ■Threat actors are actively targeting and stealing AI models, prompts, and API keys, creating new avenues for intellectual property theft and the deployment of malicious AI.
• ■Advanced AI models are demonstrating capabilities to autonomously generate sophisticated exploits, such as a Chinese AI model creating a Chrome exploit for a minimal cost.
• ■A vast majority of senior leaders expect increased cybersecurity budgets as the rollout of frontier AI models escalates the threat landscape.