Resources / Blogs
Long-form posts are still being written. In the meantime, these are the walkthroughs that explain how Silex thinks: what a blocked attack does and does not prove, how policy alternatives get ranked, and what has to hold before anything reaches production.
Start here
The scenario
A finance agent reads a poisoned email, calls a vendor-update tool, and the guardrail holds. Walk through what that block proves — and what it leaves unanswered about the same unsafe outcome.
ReadThe two jobs
Coverage asks which paths remain uncovered and which controls can still be bypassed. Optimization asks which policy change is actually best. Neither question stands on its own.
ReadHow it works
Reconstruct the environment, explore the paths that were never tested, then generate policy alternatives with rationale, expected impact, evidence, and confidence.
ReadValidation
Why no candidate policy reaches production on a simulation result alone, and what each stage of the promotion ladder is there to rule out.
ReadWhat's different
Coverage and optimization together, causal security reasoning, sim-to-real validation, business-aware recommendations, self-evolving intelligence, and a governed agentic harness.
ReadResearch
Security and trust papers on prompt injection, agent memory, poisoning, and automated red-teaming — the failure modes the engine has to reproduce on a real agent.
ReadOn the way
Join the early access list and new posts will reach you as they publish.