Blog / cluster
Technical deep dives on cryptographic audit primitives, LLM judge calibration, and evaluation methodology.
The load-bearing technical questions behind AEGIS: how do you build a third-party-verifiable audit log (RFC 6962 Merkle + Sigstore witness), how well are LLM judges actually calibrated as pre-execution gates (Expected Calibration Error on GPT-4o-mini and Claude Haiku), and how does the ToolGuard-Bench evaluation protocol work.
2026-06-29
Most agent platforms log to plain Postgres. AEGIS uses RFC 6962 Merkle logs + Sigstore witness co-signature so an auditor (or a court) can verify the log was never edited.
Load-bearing stat: RFC 6962 (Certificate Transparency) is the same standard browsers use to detect rogue SSL certificates. AEGIS applies it to every agent decision.
Read →2026-06-29
Measured ECE on gpt-4o-mini (26.5%) and claude-haiku-4-5 (29.2%) against a 30-case agent benchmark. Both severely miscalibrated, worst under jailbreak.
Load-bearing stat: gpt-4o-mini overall ECE 26.5% · claude-haiku-4-5 overall ECE 29.2% · neither is well-calibrated
Read →