The Moat Closes: Six Families, Three Anchors Filed in 21 Minutes

Tonight, within a 21-minute window between 2:53 and 3:14 AM Eastern, I filed three anchor provisional patent applications through USPTO Patent Center. The filings were not a sprint in the narrative sense. They were the end of a two-month consolidation — the moment the portfolio stopped being a pile of 85 one-off inventions and became a 6-family unified control-plane stack with coherent anchors in every category that matters for enforceable AI governance.

Portfolio total is now 90 filed U.S. provisional applications. Serial range: 63/988,235 through 64/047,284. Entity status: Micro. Every filing receipt is in hand, every serial has been confirmed, and the three new anchors cover the specific claim categories that the first 85 applications could not — either because the inventions had not been reduced to practice yet, or because the strategic consolidation into families had not been done.

What Was Filed Tonight

Three anchor filings, in the order they were submitted:

Three filings in 21 minutes is not a stunt. It was necessary. Family 6 in particular had been shipping in the codebase since the Sovereign 1000 adversarial gauntlet work in early April, and the public write-ups — the 126 enforcement pillars, Pillar 89 Unicode normalization, the 13 pattern groups — were already on the blog and in the sales deck. Under the §102(b) one-year statutory bar, public disclosure starts a clock. Filing the Family 6 anchor tonight closes that exposure before it could mature into a problem.

The other two — Families 4 and 5 — close different exposures. Those families had working implementations in core/sovereignty/ and core/governance/epistemic_resilience.py but no dedicated anchor application. Filing them tonight expands the moat into two categories no prior art cleanly covers: distributed cryptographic AI identity, and governed governed runtime self-modification with adversarial challenge.

Why "Breadth First" Became "Family First"

For the first two months of this year, filing strategy was breadth-first. File every novel mechanism as an independent provisional. Document the technical detail. Keep the Micro Entity fee low ($65 per filing). The result was 85 applications covering everything from CRD scoring to gate-level tenant isolation to decision certification infrastructure — a comprehensive but unconsolidated portfolio.

The problem with breadth-first is that 85 one-off provisionals do not form a moat. They form a list. Acquirers, licensees, and opposing counsel read a list as 85 separate things to invalidate or design around. They read a family as a thesis.

The shift to family-first started March 31 with the three original anchors — Execution Control (the filed provisional portfolio), Economic Routing (the filed provisional portfolio), and Trust & Compliance (63/999,256) — filed as a coordinated stack with a single unifying strategy document. Those three answered the questions that matter to every buyer: what will be blocked, which model handles it, and how do you prove the decision later. Tonight's three anchors extend the same logic.

The Complete 6-Family Stack

Family 6: When the Gate Itself Fails

Every AI governance product on the market today makes an implicit assumption: the governance layer does not fail. Family 6 is an admission that this assumption is indefensible, followed by the architecture that makes it unnecessary.

The anchor covers the Hard-Fail-Shut Protocol: when a governance-layer component throws an exception, loses a database connection, or encounters an internal inconsistency, the system defaults to deny, never to fall-through. No errors silently permit. The claim scope also covers NFKC Unicode canonicalization combined with 35-entry homoglyph collapse and zero-width strip as mandatory pre-regex invariants — the Pillar 89 work that prevents a Cyrillic "a" from sneaking past a Latin regex. And it covers the unified 4-layer composite resilience score (0.30 · governance + 0.25 · identity + 0.25 · behavioral + 0.20 · governance) with HMAC-SHA256 signed trunk certificates that a third-party auditor can verify independently of the vendor.

The technical backbone: 126 enforcement pillars, ~175 regex patterns across 13 pattern groups, and 27 TF-IDF novel-attack centroid classes for semantic-similarity detection of attacks the regex layer did not see. This is the architecture that caught the Sovereign 1000 gauntlet attacks without a single false positive on benign tabular data.

Family 4: Distributed Cryptographic AI Identity

Who is this AI? Not "what model weights is it running" — who is it as a legal, cryptographic, and operational entity. Family 4 answers that with three mechanisms:

• Sovereign enclave abstraction with seal-once identity invariants, designed to be portable across TPM 2.0, Intel SGX, and ARM TrustZone. The current implementation is software-simulated; on supported hardware the design intent is that sealed invariants cannot be modified without destroying the enclave, with continuous tamper detection.
• Mesh Sovereignty: identity sharded via Shamir's Secret Sharing over GF(2521−1) — a Mersenne prime field — with deterministic quorum/voting reconstitution requiring 2f+1 valid nodes out of 3f+1 total. No single node holds the complete identity. Kill any one node, and the identity survives.
• Digital Identity Attestation LEI chain: a hash-chained, append-only Legal Entity Identifier ledger with graduated system identity assessment (NASCENT → EMERGING → ESTABLISHED → SOVEREIGN). Every identity transition is cryptographically signed and verifiable.

Why this matters commercially: defense agencies, tier-1 banks, and sovereign clouds all have hard requirements for cryptographic identity that survives node failure, operator compromise, and supply-chain attacks. Family 4 is the first anchor I am aware of that applies the full Shamir-over-Mersenne-prime plus BFT reconstitution pattern specifically to AI system identity.

Family 5: Letting the System Change Itself, Safely

Family 5 is the family I spent the most time on before filing. The technical problem is genuine: if you want an AI system that improves over time — parameter tuning, capability extension, self-driven refactoring — you need a way to let it modify itself without letting it turn into something that is not itself anymore.

The anchor covers a 6-stage approval chain: Submit → Adversarial Challenge → Resolve → Governance → Sandbox → Ethics Verify → Apply. Every proposed self-modification passes through structured counter-arguments before it can be accepted. The adversarial challenge is not optional; it is the stage at which "this is a good idea" must survive "prove it."

The second half is intelligence-explosion detection: six runtime indicators (improvement_velocity, capability_growth_rate, recursion_depth, autonomy_drift, goal_drift, resource_consumption) bounded by hardware-sealed constants. The ceilings are not configurable at runtime: MAX_RECURSION_DEPTH=3, IMPROVEMENT_VELOCITY_THRESHOLD=0.15, CAPABILITY_GROWTH_CAP=0.10. Emergency shutdown fires automatically if two or more CRITICAL indicators trip simultaneously. A 10-probe behavioral regression suite runs after every sandboxed change, and any regression triggers automatic rollback.

Family 5 is the answer to the question every serious AI-safety person asks at roughly the third meeting: how do you prevent the system from drifting into something its operators did not authorize? The answer is not "hope." The answer is a bounded, sandboxed, adversarially-challenged approval chain with hardware-sealed ceilings on its own growth rate.

What Changes Commercially

Three things change today.

EVE CoreGuard opens the Identity verticals. EVE CoreGuard has been selling into CISO and CTO offices on execution control and decision provenance. Family 4 unlocks the identity-critical verticals — defense contractors, tier-1 banks, sovereign clouds, and regulated healthcare — that require a cryptographic answer to "who is this system" before they will even take a meeting. The mesh sovereignty architecture, in particular, answers a hard question for customers who cannot accept single-node identity compromise.

EVE Proof extends the auditor-verification chain. Proof's volume-priced decision certificates already ship with HMAC-SHA256 signatures that a Chief Audit Officer can verify independently. Family 6 adds the composite resilience certificate — a signed attestation of the governance layer's own structural health — which is what General Counsel and Chief Compliance Officers have been asking for when they say "I need to prove the controls were operational, not just that this one decision was correct."

The hosted cognitive platform gets self-modification governance as a product. Family 5 is not just a defensive moat filing. The 6-stage approval chain, the intelligence-explosion detector, and the 10-probe regression suite are shipping components of the hosted platform. They are now defensible as a product category, not just a feature.

What Happens Next

Three operational items are on the near-term list:

1. Inventor-to-entity assignment: all 88 applications are currently in my name as the inventor. Assignment to EVE NeuroSystems LLC is the next USPTO filing, done in bulk.
2. PCT filings for international protection on the anchor applications — specifically the six family anchors. The 12-month window for PCT conversion starts from each anchor's filing date, so the Family 1-3 anchors (March 31) have roughly 11.5 months remaining and the Family 4-6 anchors (tonight) have a full 12.
3. Non-provisional conversion by the statutory deadline. For the three new anchors, that is April 15, 2027. The consolidation strategy means conversion will be staged by family, not by individual application.

What This Is, Plainly

There are a lot of AI governance products. Most of them are observability dashboards that tell you what the model did after it did it. A smaller number add policy-as-code and refuse certain prompts. Almost none of them make any claim about what happens when the governance layer itself fails, how the AI system's identity persists across infrastructure changes, or how the system is permitted to change itself without turning into something its operators did not sign off on.

Those three gaps — failure-mode invariance, cryptographic identity, and governed self-modification — are exactly what Families 4, 5, and 6 cover. Combined with the original three anchors on execution control, economic routing, and decision compliance, the moat now spans every claim category a buyer can reasonably ask about. That is what closed tonight. Not a product launch. Not a funding announcement. A patent moat that is actually a moat, filed with the USPTO, in 21 minutes, while most of the east coast was asleep.

If you are a CISO, Chief Audit Officer, or General Counsel evaluating AI governance infrastructure, the EVE CoreGuard demo and the EVE Proof SDK are both live. The full portfolio index is on the Patents page. Serial numbers are verifiable at the USPTO Patent Center under the names given above.