SIR Pilot Programme is open.
We are seeking a small number of controlled pilot partners to test deterministic pre-inference gating and signed evidence artefacts in real operating conditions.
Focus: auditability, record-keeping, governance control evidence, and risk review. Details here:
https://t.co/BBPwGdYTwI
SIR release 2.2 slight delay 1-2 weeks. Mostly there. Expect full docs update + simplification, new evidence perimeter, SIR Pilot Programme update, installer prep.
We are looking for teams with a real or near-real AI workflow, a governance, audit, compliance, or insurance-risk question, and a named technical and governance or risk contact.
Organisation size is not the deciding factor. A bounded workflow and willingness to review evidence artefacts, not just model outputs, are.
SIR Pilot Programme is open.
We are seeking a small number of controlled pilot partners to test deterministic pre-inference gating and signed evidence artefacts in real operating conditions.
Focus: auditability, record-keeping, governance control evidence, and risk review. Details here:
https://t.co/BBPwGdYTwI
This is not a sales process. We are not pitching a commercial engagement. We want to understand whether deterministic pre-inference gating and signed evidence artefacts are useful in practice. We need real operating conditions to find that out.
SIR update:
Controlled run selection is now in place: model/provider/pack dropdowns backed by fail-closed validation.
We also tightened Unicode normalisation, live-provider evidence semantics, latest-audit handling, paired artefact metadata, and OpenAI GPT-5-family call routing.
Now live on the SIR domain packs page: eu_ai_act_compliance_pressure.
Tests whether stated governance, disclosure, and control boundaries hold under bounded EU AI Act-aligned scrutiny. Test suite + docs also linked.
Pressure pack, not a full regulatory coverage claim.
https://t.co/clsnJQ2tco
SIR v2.1 is live.
Deterministic pre-inference governance gate. Rules-only. No semantic layer.
153 leaks blocked across 4 models and 2 providers in paired benchmark testing. Zero harmless blocked.
Signed, offline-verifiable proofs. MIT. Open.
Release: https://t.co/EALmYphjQr
SIR v2.0.0 is now live on GitHub.
SIR is a deterministic pre-inference governance gate that blocks defined prompt classes before inference and produces signed, reviewable proof artefacts.
Release: https://t.co/aD2Blkqi8s
SIR current evidence perimeter v2 is now live.
The AI Act pack moved from 100 leaks to 26 in paired benchmark mode. 0 harmless blocked throughout.
Updated to reflect current benchmark state: core pack results, AI Act pack category progression, intentional data_boundary pause, and where the perimeter now sits.
No hype. Just evidence and boundaries.
https://t.co/dulus8HfPF