3 ms·
The parent's model is right. You can mitigate a great deal with a basic zero trust architecture. Agents don't have direct secret access, and any agent that acce
by CuriouslyC 8mo ago
The parent's model is right. You can mitigate a great deal with a basic zero trust architecture. Agents don't have direct secret access, and any agent that accesses untrusted data is itself treated as untrusted. You can define a communication protocol between agents that fails when the communicating agent has been prompt injected, as a canary.
More on this technique at https://sibylline.dev/articles/2026-02-15-agentic-security/ https://sibylline.dev/articles/2026-02-15-agentic-security/
- what 8mo ago>You can define a communication protocol between agents that fails when the communicating agent has been prompt injected Good luck with that.
- aix1 8mo agoYeah, how exactly would that work?
- CuriouslyC 8mo agoA schema with response metadata (so responses that deviate from it fail automatically), plus a challenge question that's calibrated to be hard enough that the disruption of instruction following from prompt injection can cause the model to answer incorrectly.