United in Defense: Architecting Safe and Trustworthy AI Agents
Abstract
Talk at BSides Seattle 2026 on architecting safe and trustworthy AI agents.
Key points:
- Adversarial robustness for multi-agent AI systems
- Defense frameworks against coordinated attacks
- Security architecture for production AI agents
- Multi-agent adversarial evaluation arenas
- Lessons from building AI systems at PayPal scale
Takeaway: As AI agents become more autonomous, adversarial testing and defense frameworks are essential — building safe AI requires thinking like an attacker.
