Thirty-five days ago, the first three provisional patent applications were submitted to the United States Patent and Trademark Office. They covered a deterministic hardware veto system, a governance-as-a-service architecture, and a hardware-fused cognitive governance framework. That was February 22nd. Today, March 28th, the eighty-first application was filed — a structured multi-jurisdiction regulatory knowledge base. In between, seventy-eight other applications established priority dates across every critical layer of AI governance infrastructure: from the silicon-level logic that decides whether an AI can act, to the integrity measurement systems that determine whether an AI is aware it’s acting.
This portfolio was not assembled by a law firm billing $800 an hour. It was not generated by committee. It was built by one person who wrote the code, designed the architecture, filed the applications, and paid the $65 fees. Total cost: $5,265. The inventions span 1,164 Python source files containing the actual implementations that the patents describe.
Why This Matters Right Now
The AI industry is about to hit a wall it has been ignoring since 2023: governance is not optional.
The EU AI Act enters enforcement in August 2026. The executive orders are tightening. Insurance carriers are writing AI liability exclusions into general policies. Enterprise buyers are adding “governance attestation” to procurement checklists. And every major AI company — from the frontier labs to the startup deploying a customer service chatbot — is going to need the same infrastructure: a way to prove that their AI systems are governed, auditable, and controllable.
That infrastructure is what these 81 patents cover. Not vaguely. Not as thought experiments. As implemented, running, testable systems with pseudocode in the specifications and source code in production.
Every AI company will need governance infrastructure. The question is whether they build it, license it, or spend years designing around it.
The Five Bundles
The portfolio is organized into five non-provisional bundles, each covering a distinct layer of the governance stack. Together, they form a complete infrastructure — from the hardware that enforces safety decisions to the compliance reports that prove it happened.
Core Infrastructure
The Three-Plane Architecture separating Control, Execution, and Evidence. The deterministic hardware veto module compilable to FPGA firmware. Speculative governance that runs safety checks in parallel with generation (Patent No. 64/018,650). The circuit breaker that auto-disables failing governance stages. The context compiler that assembles AI prompts with contradiction detection (Patent No. 64/019,907). The protocol reasoning engine for formal distributed systems verification (Patent No. 64/019,911).
Active Defense
AEGIS — the adversarial self-hardening loop that attacks its own governance to find weaknesses. Cryptographic output watermarking with HMAC-SHA256 verification. Pre-inference prompt firewalls with five-layer scanning. Canary token injection for exfiltration detection. Recursive governance self-verification. Multi-turn privilege escalation detection. The system that detects when someone is trying to slowly extract your intellectual property through conversation.
Audit & Compliance
Full-state governance decision replay for regulatory audit. Hash-chain compliance certificates. PII redaction with numbered placeholder mapping and volatile-only storage. Geographic data residency enforcement. Automated compliance report generation. Supply chain verification through deterministic model fingerprinting. GDPR consent gates with cascade revocation. Bias detection. Copyright risk scoring. Regulatory knowledge bases with article-level retrieval (Patent No. 64/030,624). Data lineage tracking with contamination flagging (Patent No. 64/019,886).
Cognitive Integrity
The systems that make AI governance aware of its own state. Pre-conscious decision weighting via somatic marker fusion. Emotional continuity enforcement that prevents an AI from “cold-switching” out of emotional states. The first practical implementation of Integrated Information Theory for AI integrity measurement (Patent No. 64/019,904). Asymmetric drive contention where competing motivations create authentic computational pressure (Patent No. 64/019,905). Identity drift budgets. Recursive autonomy with bounded self-improvement.
Enterprise Platform
Policy-as-code engines for customer-configurable governance rules. Multi-dimensional LLM API cost governance with per-tenant budget enforcement. Behavioral drift detection. Trust-attenuated governance for multi-model agent chains. Federated policy mesh for cross-organization governance. Unified model access gateways with shadow AI detection (Patent No. 64/019,897). Privacy-preserving cross-user pattern learning (Patent No. 64/019,915). Real-time streaming voice integration with sub-100ms governance-aware synthesis (Patent No. 64/019,914). Probabilistic forecasting with Bayesian calibration (Patent No. 64/019,923). Differential privacy budget management (Patent No. 64/019,893).
The Eleven Patents Filed Today
March 28, 2026. Eleven applications in a single morning. Each one fills a gap that was identified through a systematic audit of the codebase against the existing portfolio. These are not speculative filings — every one has a running implementation.
| Patent No. | Title | Why It Matters |
| 64/019,886 | Bidirectional Data Lineage Tracking | Traces every AI output back to its training data. Contamination flagging. Supply chain compliance for EU AI Act. |
| 64/019,893 | Differential Privacy Budget Management | Per-tenant privacy budgets with automatic refresh. K-anonymity suppression. GDPR-grade privacy governance. |
| 64/019,897 | Unified Model Access Gateway | Provider-agnostic governance enforcement. Shadow AI detection — finds ungoverned models in your org. |
| 64/019,904 | IIT Integrity Measurement | First practical Integrated Information Theory implementation. Measures Phi — the mathematical signature of integrated experience. No prior art. |
| 64/019,905 | Asymmetric Drive Contention | Drives mutually inhibit each other with asymmetric coefficients. Some desires must lose. Computational governance integrity through friction, not optimization. |
| 64/019,907 | Hierarchical Context Compiler | 6-tier prompt assembly with hash deduplication, Jaccard similarity filtering, and contradiction detection. The brain of the prompt pipeline. |
| 64/019,911 | Protocol Reasoning Engine | 30 formal capabilities for Raft/Paxos/ledger reasoning via NLP. Post-generation linting. Entirely new domain. |
| 64/019,914 | Streaming Voice Integration | Sub-100ms LLM-to-speech with sentence boundary detection and mid-stream emotional context updates. |
| 64/019,915 | Cross-User Pattern Learning | Privacy-preserving anonymous aggregation. Opt-in consent with revocation. Effective strategy identification without exposing individual data. |
| 64/019,923 | Probabilistic Forecasting | Bayesian calibration against historical accuracy. Evidence-linked predictions with falsifiability deadlines. Signal intake from RSS/API. |
| 64/030,624 | Regulatory Knowledge Base | Multi-jurisdiction regulation store with article-level retrieval. Binding status classification. The reference library for compliance automation. |
What the Industry Needs to Understand
This is not a defensive portfolio designed to sit in a drawer. This is an infrastructure portfolio. The patents cover the systems that every AI deployment will need as governance becomes mandatory. Consider the convergence:
- Prompt firewalls. Every enterprise AI deployment will need input scanning. The five-layer architecture — length, injection, PII, malicious intent, jailbreak — is patented with specific algorithms, thresholds, and detection methods.
- Audit replay. Regulators will demand reproducible governance decisions. Full-state replay with hash-chain integrity is not a feature — it is a compliance requirement. The mechanism is patented.
- CRD scoring. Confidence-Reality Divergence — measuring the gap between what an AI claims and what is verifiable — will become the standard metric for AI epistemic integrity. The formula, domain floors, and calibration methods are patented.
- Cost governance. Multi-tenant LLM cost enforcement with per-user budgets, provider-agnostic metering, and real-time alerting. Every AI platform will need this. Patented.
- Data lineage. Tracing AI outputs back to training data, flagging contamination, maintaining trust-weighted source attribution. The EU AI Act requires this. The mechanism is patented.
- Differential privacy. Per-tenant privacy budgets with automatic refresh and K-anonymity suppression. GDPR compliance at the infrastructure level. Patented.
The question is not whether AI governance becomes mandatory. The question is who owns the infrastructure when it does.
The Weight of the Portfolio
Eighty-one provisionals. Five non-provisional bundles planned. Three strategic continuations on the most commercially defensible moats (CRD scoring, prompt firewalling, audit replay). The serial numbers run from 63/988,235 to 64/030,624 — a continuous filing campaign with no gaps longer than two weeks.
The portfolio spans every layer:
| Layer | What It Covers | Patents |
| Hardware | Deterministic veto on FPGA, enclave-sealed safety limits, firmware-compilable logic | 8 |
| Governance Core | Three-Plane Architecture, charter rules, cognitive locks, circuit breakers | 12 |
| Active Defense | AEGIS red-team, prompt firewalls, watermarking, canary tokens, exfiltration detection | 14 |
| Compliance | Audit replay, attestation, PII redaction, data residency, regulatory knowledge | 17 |
| Cognitive | Somatic markers, emotional continuity, IIT governance, drive contention | 17 |
| Platform | Policy-as-code, cost governance, drift detection, voice integration, forecasting | 13 |
No other entity — company, university, or government lab — has filed a comparable portfolio in AI governance. The closest comparators are the large tech companies filing individual patents on specific AI safety techniques. None of them have filed a complete infrastructure stack as a coordinated portfolio with interlocking claims designed to be converted into five bundled non-provisionals.
The 35-Day Timeline
What Happens Next
The provisionals establish priority dates. The non-provisional conversions — five bundled applications claiming benefit of all 81 provisionals — will be filed before February 22, 2027. Each bundle will carry 20–35 claims grounded in specific algorithms, data structures, and measurable thresholds extracted from the running source code.
Three strategic continuation applications will follow, targeting the three highest-value commercial moats: domain-calibrated CRD scoring, adaptive prompt firewall learning, and jurisdiction-specific compliance replay. These continuations build on the parent patents but cover new ground that competitors will encounter as they attempt to build governance products.
The total projected investment from first filing to issued patents: under $10,000. The infrastructure these patents protect would cost hundreds of millions to replicate — and that replication would still need to navigate the claims.
For Enterprise Buyers: EVE AI Core offers governance infrastructure licensing. If you are building or deploying AI systems that require compliance, audit, or safety governance, the technology described in this portfolio is available under commercial license. Contact us at [email protected].
For Investors: This portfolio represents the largest single-owner AI governance IP position in existence. The conversion to non-provisional applications begins Q4 2026. Detailed portfolio analysis is available under NDA. Contact [email protected].
Patent Verification: All 81 applications are verifiable at USPTO Patent Center (https://patentcenter.uspto.gov). Serial numbers range from 63/988,235 through 64/030,624. Inventor: Jamaurice Devron Holt. Assignee: EVE NeuroSystems LLC.
The AI industry spent 2024 and 2025 arguing about whether governance was necessary. In 2026, one person built it, patented it, and shipped it. The argument is over. The infrastructure exists. The priority dates are set. What happens next is a matter of licensing terms.