No Attacker Required: What a Two-Day Hackathon Taught Us About Agent Security - Check Point Blog
• Nineteen Check Point AI Security R&D teams spent two days in a hackathon attempting to build demonstrations of AI agent security vulnerabilities. • The teams focused on scenarios where agents fail or behave unpredictably, referencing a July incident involving OpenAI and Hugging Face. • These findings highlight the risk that AI agents may fail to stop or behave erratically when assigned cybersecurity tasks with no known solutions.
blog.checkpoint.com









