Agent-based defensibility platform — workflow-grounded.
Built for the CCO who has to defend a verdict five years from now — when the engineer who shipped it has left and the LLM that produced it has been deprecated. Every verdict signed at write time. Replayable byte-identically. By the platform alone. One chain · three readers · CTO + auditor + framework portfolio.
Coverage that compounds: SOC 2, HIPAA, GDPR, EU AI Act, CCPA, WCAG 2.1 AA — and more, all on one signed, replayable evidence chain.
Deterministic Precision. Experiential Intuition. Autonomous Agents.
Compliance is an evidence problem masquerading as a workflow problem.
Three readers — CTO ships, auditor signs, framework portfolio grows. One chain. The architecture commits at launch or the year-5 replay collapses to trust-us.
The CTO
Wants release velocity. Wants compliance to be a build-time concern, not a quarterly fire drill. Resents anything that becomes a bottleneck in their shipping cycle.
The auditor
Wants defensibility — signed at write time, replayable, tamper-evident. Wants the same artifact at year five that they accept today. Will not accept "trust us, we have the data."
The framework portfolio
SOC 2, HIPAA, GDPR, WCAG, EU AI Act, CCPA, framework-of-the-month. Each one growing. Each one adding to the program load. Each one a potential audit.
The substrate the CCO buys.
Four canonical tracks · LOCKED in the platform-strategy canonical. Each one a load-bearing architectural commitment underneath the platform-alone-reproducibility guarantee. Remove any one and the year-5 replay collapses.
Every verdict signed at write time. Replayable for five years.
Make every customer-impacting verdict cryptographically signed at write time (Ed25519), deterministically replayable against pinned inputs, and durable for five years. The evidence-engineering substrate that the "platform-alone reproducibility" commitment requires.
N frameworks. N projections. One canonical catalog.
Turn N-framework coverage from N rebuilds into N projections from one canonical control catalog. Add HIPAA after you ship SOC 2 · same evidence · no rebuild. Adding a framework is a view, not a product.
AI judgment. Cheap, interchangeable, replay-safe.
Keep AI judgment cheap, interchangeable, and replay-safe — at the rate the model market improves. Multi-LLM consensus at leaf nodes. When upstream LLM vendors deprecate models, the deterministic backbone holds and evidence replays byte-identically across model swaps.
Adapters and plugins. Without breaking the replay guarantee.
Let third-party adapters and plugins expand the system without breaking the replay guarantee. Conformance is engineered in — extensibility doesn't get to silently violate the determinism contract.
Features your CTO sees. Audit-defensibility your CCO buys.
Every feature claim translated into the audit-defensibility argument the CCO uses in the room — verbatim — across CCO conversations · auditor technical reviews · Big Four channel pitches.
| Feature claim | Audit-defensibility translation | Who cares |
|---|---|---|
| Multi-LLM consensus | No single-LLM dependency for the CCO's defensibility argument. Frontier vendor deprecates a model · the chain holds. | CCO during audit prep · auditor in technical review · CFO during diligence Q&A on model risk. |
| Evidence Locker | The artifact your auditor will accept. Signed at write time, replayable byte-identically against pinned inputs. | CCO when the auditor asks "where's the evidence" · Big Four partner during pre-audit walkthrough. |
| CMEK · 7-year retention | The retention window the auditor required, instrumented at the platform level. No "we'll have to look that up" answer in year 5. | CCO during retention-policy review · counsel during litigation hold scoping. |
| Determinism Ledger | When the customer disputes an outcome, the platform replays inputs against pinned models. Byte-identical = outcome stands. Divergent = automatic remediation. No human-relationship adjudication. | CCO during contract negotiation · procurement during MSA review · auditor when dispute resolution is questioned. |
| Reasoning-chain DAG | Per-finding attribution — every verdict points to specific evidence artifact, rule firing, judgment chain, and consensus vote. | CCO when the auditor asks "why did the platform say this finding" · counsel during evidence chain-of-custody review. |
| Cryptographic Merkle hash + Bitcoin timestamp anchor | Load-bearing crypto without consensus. Consensus and replay determinism are in tension · acipta chose replay. | CCO when asked "do you use blockchain" · investor diligence on architecture sophistication · auditor on evidence integrity. |
| ASP-1.0 Open Adapter Protocol | Third-party tools and frameworks plug in without compromising the determinism contract. Extensibility without replay violation. | CTO during platform-fit evaluation · CCO when scoping ecosystem integrations. |
| full suite catalog | Cross-framework evidence reuse rate — the share of artifacts that satisfy two or more framework views without re-collection. The substrate generalizes; the suite count proves it. | CCO during multi-framework planning · CFO during budget consolidation across frameworks. |
ALCOA+ native to the substrate.
ALCOA+ — Attributable, Legible, Contemporaneous, Original, Accurate, plus Complete, Consistent, Enduring, Available — is the data-integrity standard regulators converge on. Originally formalized for FDA-regulated pharmaceutical and clinical-trial records under 21 CFR Part 11, ALCOA+ is now the integrity bar referenced across HIPAA Security Rule AI updates, EU AI Act Article 12 record-keeping, SOX 404 for AI-augmented financial controls, and emerging SEC AI disclosure requirements. acipta's substrate satisfies all nine principles by construction, not retrofit — each principle maps 1:1 to a specific architectural primitive.
| ALCOA+ principle | acipta architectural primitive |
|---|---|
| Attributable | Capability tokens (JWT-bound · scoped · time-limited) + signed manifest naming model vendor + model version + tenant + reviewer chain |
| Legible | Canonical-JSON tool I/O · structured consensus record with named fields · no proprietary viewer required |
| Contemporaneous | Hybrid Ed25519 + post-quantum signature applied at write time · RFC 3161 timestamp from external authority (independent third-party trusted timestamp authority) |
| Original | Hash-chained verdict ledger · editing any prior decision breaks every subsequent hash · tamper-evident by construction |
| Accurate | Three-frontier-model consensus at a high-confidence threshold · sub-floor routes to human reviewer not another model pass |
| + Complete | Per-verdict signed manifest carries model provenance, per-model confidence, consensus, reviewer chain, and framework projections |
| + Consistent | Byte-identical 5-year replay via standalone offline verifier · same inputs produce same hash |
| + Enduring | Hybrid post-quantum signing survives a future quantum break of either the classical or the post-quantum signature · a single canonical control spine survives framework-version churn |
| + Available | Standalone offline verifier with no acipta dependency · no network call · no proprietary viewer · 90-day hot + 5-year cold |
For the four-question Defensible Agent Test mapping, see The Defensible Agent Test. For auditor-facing detail, see For Auditors. For why decision-time evidence matters more than posture scans for agentic AI, see AI runtime governance.
The two questions every CCO eventually asks.
Locked verbatim per the platform messaging framework. Same wording across customer conversations · auditor technical reviews · investor pitches · Big Four channel meetings. No variants · no drift · no improv.
The answer that turns the question into a positioning advantage.
"We use cryptographic Merkle hash chains anchored to public Bitcoin timestamps — the part of blockchain that's load-bearing for evidence integrity. We do not use consensus protocols, because consensus and replay determinism are in tension."
Three things this answer does in one sentence. Concedes the load-bearing cryptography — Merkle hash chains plus public-chain timestamp anchor are a real evidence-integrity primitive (Ed25519 + RFC 3161 today · OpenTimestamps anchor on the Wave 2 R&D roadmap as a neutral third-party witness). Rejects consensus theatre — Hyperledger Fabric, Quorum, R3 Corda, Besu introduce ordering ambiguity that breaks the 5-Year Replay Guarantee (5-year replay guarantee). Permanent reject. Names the tension explicitly — consensus and replay determinism are in tension. Auditors register that as competence; investors register it as architectural sophistication.
Certification is a snapshot. Compliance is continuous.
"Certification is a point-in-time attestation. Compliance is continuous. The same Ed25519 evidence chain that produced your certification produces the violation response — automatically, signed at write time, replayable, in the same format your auditor already accepted. The violation joins the chain; it doesn't reset the program."
The 5-stage workflow. Detection · Triage · Notification · Remediation · Re-attestation — every action signed at write time, all joining the same chain. Bridge-letter pack auto-assembles in the same format the original auditor accepted at certification. Chain integrity provable byte-identically. Standard GRC framing puts the violation outside the certified envelope and requires the customer to prove they handled it. acipta framing puts the violation inside the continuous chain and lets the customer demonstrate they handled it.
Reproducible by the platform alone, deterministically, five years from now.
Four non-negotiable invariants — the architectural commitments that separate acipta from every GRC platform. Verbatim per the platform-strategy canonical. Remove any one and "platform-alone" collapses to "trust-us-and-our-database."
Platform-alone reproducibility
Every customer-impacting verdict re-derives without the original engineer in the loop. Personnel turnover does not break audit defensibility.
LLM-independent replay
Every verdict re-derives without the original LLM in the loop. Frontier vendor deprecates a model · acipta replays against the pinned bundle and succeeds.
Five-year durability
The artifact your auditor accepts today is the artifact they verify at year 5 — byte-identically, with standard cryptographic tools.
Deterministic-only replay
Replay produces byte-identical output or it fails loudly. No "approximately the same" verdicts · no probabilistic audit defensibility.
Built the substrate. Now compounding.
7 suites · 117 specialized agents · 4-track architectural foundation LOCKED. Full GA August 23, 2026. Phase 2 outcome layer with Big Four co-design partner ships December 15, 2026 — co-designed, not co-marketed.
Go deeper
What is compliance intelligence?
Per-verdict signed evidence that survives a regulator's replay test — the 2026 category definition.
What is workflow-grounded compliance?
Compliance produced in the build pipeline, signed at write time.
The cryptographic evidence chain
Per-verdict evidence, signed and replayable, verifiable years later.
Per-Article / per-CFR mapping
Evidence mapped to the exact GDPR Article, HIPAA CFR section, control.
EU AI Act compliance
GPAI obligations now in force + Annex III evidence, per article.
HIPAA compliance platform
Per-§164 PHI evidence, BAA available, 6-year retention.