Technical writing on AI agent memory, security, and compliance.
Foundations
Why memory governance matters, and how Sgraal approaches it.
AI agents act on memory. We thought someone should decide if that memory is safe to act on. Memory governance is the missing layer between storage and observability — here’s why it matters now, and what Sgraal does about it.
Read more →Research
Adversarial corpora, scoring engineering, and joint benchmarks.
When we wrote the Round 12 ceiling analysis three weeks ago, we predicted 54/60. We were wrong. This is the honest accounting.
Read more →Seven rounds of adversarial memory testing with Grok — 554 cases, 100% recall on both separately-built stacks (synthetic adversarial corpora; production calibration pending) — convergent, but shared-LLM-prior corroboration, not independent validation. A practical epistemology for AI memory governance.
Read more →The same math that predicts equipment failure in aerospace and nuclear now predicts AI agent memory failure.
Read more →The first application of the circuit breaker pattern to AI agent memory governance.
Read more →How Sgraal and Grok scored the same adversarial memory corpora side-by-side — 100% recall across 239 cases against Sgraal's published ground truth (synthetic corpora; production calibration pending).
Read more →F1 measured against Sgraal's own ground truth on synthetic corpora; Grok independently scored the same corpora — corroboration, not external validation. Production calibration pending.