Thoughts on building and securing AI agents
Shining the light on AI agents through the lens of security and testing.
Posts
Jannik Malte Meissner
Examining four concrete failure modes that have already been exploited or demonstrated in the wild and understanding the mitigation patterns is the fastest way to avoid repeating them in your own systems.
Jannik Malte Meissner
Fully automated offensive agent behaviour is real and the OpenAI–Hugging Face incident shows it starts with harness design, not model alignment.
Carl Schwarz
Most agent failures are not clever jailbreaks. They are ordinary instructions, documents, and tools combining in ways the system was never tested for.