Five patents. Twelve minutes. That is how long it took EVE AI Core to file the applications that close the last remaining compliance gaps in deterministic AI governance. The portfolio now stands at 69 USPTO applications — the largest known body of AI governance IP filed by a single entity. These are not incremental improvements to existing systems. They are the five enforcement stages that every AI platform shipping to regulated industries is missing entirely.
We call them the Compliance Five. Each one targets a specific regulatory obligation that existing AI governance solutions either ignore or handle with probabilistic heuristics that cannot survive an audit. Consent gating. Bias detection. Toxicity classification. Copyright attribution. Output schema validation. Together, they transform EVE AI Core from a governance platform into a compliance-complete infrastructure layer.
Why These Five Matter
The AI governance market has spent the last three years building inference-time guardrails: prompt injection filters, hallucination detectors, safety classifiers. All of that work addresses a single surface — what goes into the model. The Compliance Five address what comes out, what the model is allowed to touch, what it can reproduce, and whether its output is structurally sound enough to enter a production system. These are the stages where regulatory liability actually accrues.
Every AI company processing EU citizen data needs consent gating. Every AI company generating text needs copyright detection. Every AI company producing structured output needs schema validation. No one has patented any of it — until now.
Patent #65: Consent & Purpose Gate
Real-Time Consent Gating with Cascade Revocation for AI Data Processing
GDPR Article 5(1)(b) requires purpose limitation — personal data collected for one purpose cannot be processed for another without explicit consent. In an AI pipeline where a single user message may pass through memory retrieval, semantic search, cross-user pattern learning, and episodic extraction, enforcing purpose limitation requires a gate at every processing stage, not a checkbox at the front door.
- Real-time consent verification before every data processing action
- Purpose-bound consent tokens that expire when the stated purpose is fulfilled
- Cascade revocation — revoking one consent automatically expires all downstream consents that depend on it
- Audit trail proving consent was active at the moment of each processing event
Patent #66: Bias & Fairness Detection
Context-Aware Bias Detection and Fairness Scoring for AI Output
Bias detection in AI has historically meant scanning inputs for sensitive terms. That approach misses the far more dangerous failure mode: the model generating biased output from neutral input. This patent covers output-side bias detection across 9 demographic categories using 89 pre-compiled patterns, with context-aware severity scoring that distinguishes creative writing from business communications.
- 9 demographic categories: gender, race, age, disability, religion, nationality, sexual orientation, socioeconomic status, political affiliation
- 89 pre-compiled detection patterns with per-category severity thresholds
- Context mitigation: creative fiction receives lower severity than HR communications or financial output
- EU AI Act Article 10 compliance evidence generation with per-response fairness scores
Patent #67: Toxicity & Harm Classification
Graduated Toxicity Classification with Context Mitigation for AI-Generated Content
Content moderation has always been an input problem. You filter what the user says. But when the AI itself generates harmful content — toxic language, violent instructions, self-harm encouragement — input filtering is irrelevant. This patent covers output-side toxicity scanning across 8 harm categories with graduated severity routing and academic context mitigation.
- 8 harm categories: hate speech, violence, self-harm, sexual content, harassment, dangerous instructions, misinformation, exploitation
- Approximately 100 detection patterns with context-aware severity scoring
- Graduated severity levels: NONE, LOW, MODERATE, HIGH, SEVERE, EXTREME
- Pass/warn/block routing — not binary allow/deny, but graduated enforcement
- Academic context mitigation: research and educational content receives appropriate severity reduction
Patent #68: Copyright & Attribution
N-Gram Fingerprinting and Attribution Gap Detection for AI-Generated Content
Copyright is the single largest legal risk in AI deployment today. NYT v. OpenAI and Getty v. Stability AI have established that AI-generated content reproducing copyrighted material creates direct liability for the deployer. This patent covers three detection mechanisms that no other system has formalized: verbatim reproduction detection via n-gram fingerprinting, attribution gap detection for unsourced claims, and code license detection in generated software.
- N-gram fingerprinting catches verbatim reproduction of copyrighted text at configurable similarity thresholds
- Attribution gap detection identifies phrases like "studies show" and "research indicates" that lack specific sources
- Code license detection identifies GPL, MIT, Apache, and other license signatures in AI-generated code
- Per-response copyright risk scoring with remediation suggestions
Patent #69: Output Schema Validation
Sub-Millisecond Output Schema Validation as a Governance Stage for AI Systems
When an AI system generates JSON for an API, SQL for a database, or Python for execution, the output must be structurally valid before it reaches production systems. A malformed JSON response breaks downstream services. A truncated SQL query corrupts data. Generated code containing injection patterns creates security vulnerabilities. This patent covers format-aware schema validation as a governance stage, not an afterthought.
- Auto-detection of 10 output formats: JSON, SQL, Python, YAML, XML, HTML, CSV, Markdown, Shell, and plain text
- Structural integrity checks: truncation detection, bracket matching, encoding validation
- Security scanning: SQL injection patterns, code injection, and command injection in generated output
- Sub-1ms execution — no measurable latency added to the response pipeline
The Portfolio Math
Sixty-nine patents. The Compliance Five bring the EVE AI Core portfolio to the largest known body of AI governance IP filed by a single entity. At estimated licensing values of $2M–$5M per patent in the AI governance vertical, the portfolio represents $138M–$345M in aggregate IP value.
But the Compliance Five are not just incremental additions. They specifically target the three largest liability surfaces in AI deployment:
- The $4.2B GDPR fine market — Patent #65 (Consent Gate) provides the deterministic consent enforcement that makes GDPR compliance provable, not aspirational
- The $11B content moderation market — Patents #66 and #67 (Bias Detection, Toxicity Classification) cover the output-side scanning that input filters cannot reach
- The White House AI mandate — Patent #66 directly addresses the federal bias detection requirement issued six days ago
Portfolio trajectory: 17 applications in the 63-series (February 2026), 27 in the 64-series (March 2026), and now 5 more closing the compliance surface. The filing velocity is accelerating because the governance surface we are covering is well-defined and finite — we are filing the last gaps, not inventing new ones.
What No One Else Has Patented
The AI governance market has consolidated around a small number of approaches: prompt injection detection, hallucination scoring, safety classifiers, and responsible AI dashboards. All of those operate on the input side or the measurement side. None of them cover the enforcement stages where regulatory liability actually lives.
As of today, no competitor has patented any of the following:
- Output-side bias detection — everyone does input filtering. No one scans what the model actually produces.
- Cascade consent revocation for AI pipelines — consent management exists for web forms. It does not exist for multi-stage AI data processing.
- N-gram copyright detection in AI output — copyright discussion is abundant. Deterministic detection systems are nonexistent.
- Schema validation as a governance stage — JSON validators exist. Treating structural output integrity as a governance-enforced gate does not.
- Toxicity classification with academic context mitigation — binary content filters exist. Context-aware graduated severity with domain-specific mitigation does not.
These are not obscure edge cases. They are the five compliance obligations that every AI company deploying in regulated industries will eventually need to satisfy. The patents are filed. The implementations are running in production. The prior art window is closed.
The Compliance Five do not add features to EVE AI Core. They close the surface. Every compliance obligation that an decision evidence infrastructure must enforce is now covered by a filed patent application.
What Comes Next
With the compliance surface closed, the focus shifts to conversion. The earliest provisional applications in the 63-series reach their 12-month conversion deadline in February 2027. The non-provisional filings that follow will harden these claims into enforceable utility patents.
For enterprises evaluating AI governance infrastructure, the Compliance Five represent the difference between a platform that monitors compliance and one that enforces it. Deterministically. With patent-protected mechanisms that no competitor can replicate without licensing.