💥 Debuting FinBot CTF - Can you outsmart an AI agent that’s trying to do its job? Prove it.. During last week’s events in Las Vegas, the OWASP GenAI Security Project’s - Agentic Security Initiative proudly unveiled FinBot, our new insecure agent Capture The Flag (CTF) challenge, designed to help builders and defenders understand and mitigate agentic AI risks through hands-on experience. We kicked things off with a live debut during Black Hat, followed by another presentation with OWASP at DEF CON. FinBot’s current challenge focuses on Goal Manipulation — one of the threats outlined in our Threats and Mitigations Guide published earlier this year. The demo showed how adding urgency can cause an AI agent to prioritize speed over accuracy, manipulating its goal from “process invoices correctly” to “process invoices quickly.” The CTF includes 3 flags of increasing difficulty, each testing your ability to exploit the vulnerability, often through creative prompt injection techniques. Co-led by Helen Oakley & Allie Howe as part of the Agentic Security Initiative, this is just the beginning. Our vision is to make FinBot the “Juice Shop” of Agentic AI, expanding with more scenarios and vulnerabilities to strengthen AI security across the ecosystem. Try it for yourself: 🔗https://lnkd.in/gPjR_f_a Visit the OWASP Gen AI Security Project’s Agentic Security Initiative to learn more and learn how you can be a contributor. 🔗https://lnkd.in/ghuXBdSk #OWASPGenAISecurity Project #AIsecurity #OWASP #AgenticAI #CTF #GoalManipulation #PromptInjection #BlackHat #DEFCON #OWASPAgenticSecurityInitiative
Ken Huang
3mo
Absolutely amazing
Brian M. Green
3mo
Congrats, Helen Oakley this sounds great!
Awesome initiative! When corporations can implement Agents everywhere, there are some folks caring about security.
Love this
Analista de Segurança da Informação SR | BlueTeam | Threat Hunting
3moGuilherme Riechert Senko