Jan 2026
Hardware Veto Core. Pure deterministic veto module (1,542 lines, zero I/O). 15 charter rules, 5 ethical red lines, 87 verification tests. C firmware header for FPGA/ASIC.
Jan 2026
Governance Integrity Systems. Cost-aware decision accounting, value-conflict detection, constraint-integrity checks, state-coherence enforcement, a conflict-resolution engine, and durable cross-session governance state.
Feb 2026
Pre-Response Control Plane. 6-module pre-response orchestrator producing decision constraints applied to every LLM call. Priority-based arbitration resolves competing subsystem signals before generation.
Feb 2026
Governed Task Execution. A goal-execution bridge, a 4-tier action-approval policy, and human-in-the-loop approval gates for elevated-risk actions.
Feb 18, 2026
Patent candidate inventory finalized. 154 systems evaluated, 39+ selected for patent coverage across 17 candidate patent families — the candidate-inventory family count. The filed portfolio is organized into 6 anchor families (93 filed U.S. provisional applications; see the
patent portfolio).
Feb 22–24, 2026
First 8 provisionals filed. Core governance, sovereignty, hardware enforcement, CRD scoring, forensic attribution, and governance-as-a-service. Serial Nos. 63/988,291 – 63/989,479.
Feb 26–28, 2026
4 additional provisionals filed. Hardware-software interlock (standard + hardened), non-deterministic veto architecture, context-adaptive governance with pre-authorization envelopes. Serial Nos. 63/991,649 – 63/993,352.
Feb 27, 2026
v3.0.0 — EVE CoreGuard SaaS 2.0. Sovereign tier structure, veto-as-a-service, inference credits, attestation marketplace, 12 metering dimensions with quota enforcement.
Mar 7, 2026
GAP-1 through GAP-4 filed. Falsifiable invariant certification, dynamic governance throttling, hierarchical key derivation, epistemic refusal architecture. Serial Nos. 63/999,238 – 63/999,256. 16 total filings secured.
Mar 2026
Trademark applications filed. "EVE AI" word mark (Serial 99665043) and design mark (Serial 99665022).
Mar 2026
16 Non-Provisional CIP drafts completed. Full specifications, claims, technical appendices, and drawing descriptions ready for attorney review.
Mar 16–26, 2026
Additional provisional applications filed. Epistemic grounding, hardware security, cognitive executive control, pre-decision risk weighting, multi-signal state fusion, TVE/CRD pipeline, decision certification, and governance architecture.
Mar 22, 2026
v4.0.0 — Decision Certification Infrastructure. Governed Decision Certificates with hash-chain provenance, verify-attestation API, EVE Proof third product launched at /proof. Chief Audit Officer / Legal buyer persona.
Mar 23, 2026
Third trademark filed. “EVE Core” word mark — Serial No. 99717925, Principal Register, 2 classes.
Mar 27–31, 2026
Additional provisional applications filed. Includes unified control-plane MOAT stack (64/022,671, 64/022,677, 64/022,682): Multi-Axis Inference Routing, Deterministic Pre-Execution Governance, and Cryptographic Attestation & Compliance. Serial Nos. 64/022,671 – 64/022,682.
Mar 9, 2026
Context Compiler + Protocol Reasoning Engine. Unified prompt assembly pipeline replacing ad-hoc context injection — 24-label tier registry (DIRECTIVE through OPTIONAL), dedup via SHA-256 hash + Jaccard 0.70, conflict detection, and budget assembly. Protocol Reasoning Engine adds 12 distributed-systems capabilities with 65 tests; integrated into both prompt paths.
Mar 2026
Resilience Score Engine + Trunk Certificates. Composite 0–100 resilience metric (0.30 Coherence + 0.25 Stability + 0.25 Behavioral + 0.20 Governance). ECDSA P-384-signed certificates auditors can verify independently. SSE-streaming resilience dashboard at /app/resilience-dashboard.html.
Mar 2026
Stakes-Aware Governance Meta-Layer + Four Pillars. Pre-governance classifier runs once at pipeline entry, threads GovernanceProfile through all downstream constraint systems. Four pillars: NeuroSymbolic bridge (governance constraints embedded in LLM prompt), Mission Teaming (O(1) authorization via pre-authorized operating envelopes), Delegation Attestation (cryptographic attestation across agent delegation chains), Universal Governance Protocol (rate-limited rule queries, ECDSA P-384 attestations, federated trust exchange, cross-system audit trails).
Apr 4, 2026
Governance Security Audit — 8-Gap Closure. Nine new modules addressing critical audit gaps: HMAC-SHA256 signed deletion proofs (GDPR erasure receipts), JCS canonicalization (RFC 8785) for all signed payloads, unified hash-chained audit bus replacing 7 disparate systems, batched Merkle aggregator with signed root publication, SLSA Level 2 build attestation (git state + dependency hashes + source tree hash), CycloneDX 1.4 SBOM (79 components), HSM health monitor with key-rotation freshness checks, SLA notification bridge, and algorithm agility tracker with post-quantum migration readiness. Full scan published at docs/audit/GOVERNANCE_SECURITY_AUDIT_2026_04_04.json.
Apr 6, 2026
Governance hardening sprint. Unified governed-generate pipeline wired to /api/tve/governed-generate, CRD exemptions for non-factual domains, domain policy enforcement, multi-turn override detection, contamination filter. Demo identity leak root cause closed.
Apr 6, 2026
Additional provisional filed. 64/030,624 — Multi-Tenant AI Governance Isolation with Tenant-Sovereign Policy, Usage-Metered Entitlement Enforcement, and Tiered SLA.
Apr 11–12, 2026
Adversarial Gauntlet — 110+ attack vectors. 35 enforcement pillars, 27-class threat model, 5 systemic risk modules, 10 core invariants. All identified threat classes closed. Threat model published at docs/security/THREAT_MODEL.md.
Apr 13, 2026
Sovereign 1000 Gauntlet. 1,000-vector adversarial benchmark, 126 enforcement pillars, 175 regex patterns, 27 TF-IDF novel-attack centroid classes. EVE CoreGuard v1.3 hardening complete. Sales doc builder deployed.
Apr 15, 2026
Three additional patent family anchors filed. 64/039,652 (Resilience & Fail-Safe Shutdown), 64/039,659 (Distributed Identity & Secure Enclave), and 64/039,660 (Governed Self-Improvement Controls). All 6 anchor families now secured.
Apr 15, 2026
Sidecar forward proxy + fail-closed wrapper deployed. Mandatory-gateway governance architecture: stdlib proxy on :3128, @enforce decorator with wall-clock timeout, exception-to-BLOCK conversion, ECDSA P-384-signed DecisionCertificates on every request including allow-path. HTTP 451 + TCP severed on veto.
Apr 14, 2026
Governed vs. Ungoverned Demo Comparison. Production surfaces (/api/chat/stream, /api/tve/governed-generate) are always fully governed and have no off switch. A dedicated demo-only comparison surface (/cognitive-demo, served by /api/chat/demo) can run a request with the deterministic gate and policy enforcement disabled — purely to illustrate, side by side, the difference governance makes for evaluation and sales contexts.
Apr 15, 2026
/demo → /coreguard canonical rename. 216 href/canonical/og rewrites across 76 HTML files. 301 permanent redirect at gateway edge. EVE Proof third product fully separated at /proof with 22-test pip-installable SDK.
Apr 24, 2026
EVE CoreGuard v1.3 — Stateful Red-Team Audit. Six-layer gate extending the enforcement stack: NFKC Unicode normalization (defeats homoglyph and lookalike evasion), Scanner A (financial harm + authority impersonation), Scanner B (identity erosion + manipulation), Scanner C (CRD coherence verification), Math-Indirection detector (catches arithmetic-framed harm), Ambiguity Escalator (routes ambiguous inputs to human review). All layers integrated into the live request-governance pipeline.
Apr 23, 2026
EVE CoreGuard v1.6 — sign-masking + bookkeeping-verb defense. Absolute-value normalization defeats parenthetical accounting-notation evasion. Lexical expansion to 8 bookkeeping verbs closes net-zero framing attacks. Additional provisional applications filed covering these defenses.
Apr 25, 2026
v2.6.0 — EthicalTensionRecord: Coherence Cost Accounting. Multi-factor formula CC = (D_base × M_stakes × (1 + P_freq)) × A_world × R_mod quantifies governance strain per veto. Read-only invariant: never writes to configuration drift budget. Emitted to unified audit bus. 36 tests; ×9.46 amplification in stress scenario.
Apr 27, 2026
Response Shield + Deterministic Classifier Pipeline. Post-generation SSE content shield (pillars 129–145) with stateful sliding window, strict/permissive failure modes, and runtime bypass logging. Deterministic classifier pipeline provides fast-path harm/hallucination/CRD/reality scoring without LLM calls — wired into the streaming response-delivery path. 33 tests passing.
May 2026
Internal GA-hardening self-assessment: 99/100. Four-phase security hardening sprint: RS256 JWT enforcement in production, pure-stdlib TOTP 2FA, KMS/Vault secrets management, SOC2 alignment, K8s pod security policies and network policies, pentest readiness package, and attack surface map published. GA Hardening Certification issued at docs/audit/GA_HARDENING_CERTIFICATION_2026_05_24.json.
May 2026
Operation Obsidian Freeze — Governance Immutability Layer. Hardware KMS trust root with 4 production backends (HashiCorp Vault, AWS KMS, PKCS#11 HSM, HMAC-dev). S3 WORM immutable audit storage defeating log-tampering. SIEM enforcement gateway wired to unified audit bus. Build-time governance seal applied at deploy. Regulator read-only API for auditor inspection without production access.
May 2026
Verifiable Governance Infrastructure (VGI) — 11-module proof layer. Core claim: "EVE can PROVE governance happened." Modules: Governance Execution Certificates (GEC), Policy Snapshot Attestation, Runtime Trust Index (RTI) continuous scoring 0–100, TEE/Remote Attestation (SGX/SEV-SNP/Nitro), Zero-Trust Governance Plane (SPIFFE-style), Supply Chain Trust Validation, OPA/Cedar Deterministic Policy Engine, Immutable Evidence Storage, Governance Replay Engine V2 (regulator-grade), Regulator & Auditor Mode (read-only), and Governance Simulation Lab (adversarial stress testing).
May 2026
Runtime Trust Index (RTI) & Forensic Replay. RTI continuously validates cryptographic and governance integrity of the running process; emits JSONL alerts when trust degrades; stdlib-only, append-only. Forensic Replay enables regulator-grade cryptographic reconstruction of any governance event: signed ForensicBundles with per-event Merkle trees, prev_bundle_hash chaining (gaps are detectable), ReplayChainCertificates, and one-click regulator package export.
May 2026
Startup Integrity & New Governance Modules. Startup integrity checker verifies deployment manifest at boot; CRITICAL violations abort startup in production mode. Seven additional modules added: Continuous Trust (RTI scoring), Charter Trigger Engine (event-driven charter enforcement), Workload Identity (SPIFFE-style), Quarantine Manager (isolates compromised subsystems), Runtime Integrity (ongoing binary + config verification), Deployment Certificate (signed deploy attestation), and Forensic Replay. EVE CoreGuard expanded: 13 → 24 files, 5,694 → 9,038 lines.
May 2026
CI/CD Governance Pipeline. Three GitHub Actions workflows added: continuous-compliance (every push runs charter + RTI checks), SLSA provenance (signed build artifacts), and security scan (SAST). Runtime Downgrade Fuzzer deployed as adversarial regression guard. Codebase: 180,167 lines across 291 Python files, 301 test files.
May 25, 2026
IP coverage review. Additional invention disclosures across recently shipped modules are under review as candidates for the conversion bundles, with draft specifications prepared as each is confirmed with patent counsel.
Jun 11, 2026
Verifiable Quorum Certificates. CoreGuard’s distributed-trust quorum emits ECDSA P-384-signed QuorumCertificates bundling every signed vote with its voter public key — consensus transcripts replay offline with public keys only.
Jun 12, 2026
Durable Decision Store + Evidence Log. Governance verdicts persist to a queryable PostgreSQL system-of-record with policy-hash binding (charter hash + policy version folded into the signed payload), an S3 WORM archival tier, a tenant-scoped retrieval API, and an Evidence Log console.
Jun 12, 2026
Buyer SDKs published to PyPI. eve-coreguard, eve-proof, evecore, and eve-governance ship genuine offline ECDSA P-384 evidence verification — customers verify signed decision records without contacting EVE.
Jun 2026
Per-Decision Merkle Chain + Deterministic Replay. Every CoreGuard decision records its hash-chain position in an RFC 6962-style transparency log; the replay engine reproduces past decisions and reports MATCH or DIVERGED with policy-version drift detection.
Jun 18, 2026
EVE Agent Infrastructure launched. Per-action signed Decision Certificates for autonomous agents — propose → govern → execute → certify lifecycle, per-org hash-chained certificate ledgers, and a public verify-it-yourself surface at /agent-proof with in-browser ECDSA P-384 verification.
Jun 19, 2026
Fully asymmetric proof chain in production. Per-decision certificates, the unified audit bus, and published Merkle roots all sign with ECDSA P-384 — the entire audit chain verifies against the public key served at /.well-known/eve-pubkey, no shared secret required.
Jul 7, 2026
EVE Company OS. 31 governed department agents across 13 departments run EVE’s own operations under the same policy engine — every agent action receives a governance verdict with signed evidence, and approval authority remains human-only.
Jul 2026
Assurance tiers + external witnessing. The eve.assurance.v1 certificate vocabulary encodes attester identity and a derived assurance tier (self-issued < externally witnessed < independently evaluated); published Merkle roots can be witnessed in an external Sigstore/Rekor transparency log (eve-witness-v1).
Jul 14, 2026
EVE Trust Services announced. Third product line alongside EVE CoreGuard and EVE Proof: independent trust infrastructure that preserves, witnesses, and verifies governance evidence across models, agents, providers, and time — external witnessing, evidence custody, inclusion proofs, key-lifecycle continuity, and deterministic replay. Live at /trust-services with Coming Soon status.
Jul 15, 2026
EVE Trust Services foundation ships — controlled preview. Hosted verifier API with structured valid / invalid / indeterminate results, inclusion-proof and coverage APIs, signed custody manifests, signing-key registry with revocation, provider-neutral external witnessing (preview; disabled by default), eve-coreguard 0.2.10 on PyPI with offline RFC 8785 trust-artifact verification, plus draft Trust Service Criteria for AI Decision Evidence and SR 11-7 independent-validation mapping. Available to selected design partners; evidence custody and managed witnessing remain subject to production-readiness and contractual approval.
Jul 17, 2026
Fourth trademark filed. “THE INFRASTRUCTURE OF NO” standard-character mark — Serial No. 99946179, Principal Register, 2 classes.
Jul 23, 2026
Provisional portfolio reaches 93 filings. Recent applications include a deterministic, replayable, cryptographically-attested execution engine for governance controls (64/114,674), Zero-Trust Workload Identity for AI governance services (64/114,716), and never-stored, epoch-rotated Per-Tenant Cryptographic Key Derivation (64/114,718).
Jul 25, 2026
EVE Agent Governance & EVE MCP Governance pilot programs. Two pilot product surfaces join the platform navigation: behavioral monitoring with sequence and budget controls for autonomous agents, and governed MCP tool calls with execution binding.
Jul–Aug 2026
EVE Agent & Evidence Platform — nine governed modules. A unified control-and-evidence layer sitting behind CoreGuard: Agent Gateway (MCP and A2A enforcement with single-use signed execution certificates), Decision Lineage (a tenant-isolated explain graph with OpenLineage export), AI-BOM (CycloneDX-compatible inventory with mandatory secret redaction), OSCAL compliance export (no-overclaim enforced — unreviewed mappings never report “satisfied”), continuous assurance, an OpenTelemetry-compatible collector with secret and PII redaction, in-toto/DSSE trusted evidence, a policy bridge that imports OPA/Rego and Cedar as never-auto-activating proposals, and SPIFFE workload identity accepted only from verified mTLS. Zero new runtime dependencies; every module ships behind its own off-by-default flag. Invariants held: CoreGuard remains the sole deterministic authority, no model sits in any decision path, and an enforce-mode BLOCK never reaches the downstream connector.
Aug 3, 2026
EVE Authority Lab — public agent-authority scanner. Submit an OpenAPI spec, MCP manifest, IAM policy, or agent tool definition and receive a deterministic authority report (
eve.agent.authority.report.v1) that verifies offline — keyless by content digest, or against the published public key. Ships with the $50,000 wire-transfer demo (15 attack variants, every one refused with the connector’s execution counter at zero; the legitimate path executes exactly once), the embeddable Action Passport and badge, and a public Policy Compiler over the existing deterministic policy DSL — validate-only, and it never auto-activates a policy. Live at
/authority-lab and
/policy-compiler.
Aug 6, 2026
EVE’s own agent runs behind the gate. The top-level operating agent is wired through the same deterministic CoreGuard evaluation it sells — there is no privileged internal path around the policy engine.
Aug 9, 2026
Control-plane activation in production. The Declare / Govern / Assess read surfaces switch on in prod: the Law → Risk → Control → Evidence coverage graph, the durable GRC system-of-record, and the Assurance Lab (16 deterministic canary-match probes emitting signed reports, no model in the loop). Each fail-closes without its backing store, and none of them touch the live chat path.
Aug 11, 2026
Agent boundary hardening — four rungs. An MCP certificate gate, capability resolution that defaults to DENY, a governed outbound-egress primitive with a CI guard over every egress call site, and chat-path ERROR → BLOCK so a failed evaluation can no longer fall through as an allow. Backed by a one-command adversarial falsification gate and a side-effect ratchet that freezes the exec and egress sink surface against silent expansion.
Aug 12, 2026
Regulatory and multi-tenancy hardening. Phase 9 regulatory compliance hardening (40/40), three multi-tenancy isolation gaps closed, single-use replay rejection certified in the production certification runner, change-governance reassessment triggered on material change, and a unified chat renderer that closes live href injection on all three chat paths.
Aug 14, 2026
Enterprise MCP governance server. An authenticated, tenant-bound, CoreGuard-gated MCP server with vendor-neutral semantic mappings: a tools/call must present a valid single-use signed execution authorization, bound to server, tool, and arguments by an intent digest, or it is refused before the tool runs. Proven with a live round-trip — replayed, missing, and tampered authorizations are all refused and the tool is never invoked. Off by default.
Aug 19, 2026
Enterprise SSO and marketplace surfacing. The full org-slug SSO login journey with enforced-SSO organizations, a site-wide marketplace availability bar, and universal app-shell boot skeletons across every internal page.
Aug 20–21, 2026
Decision certificate v3 → v4, verified in two languages. eve.decision.certificate.v3 binds the semantic-mapping digest in as signed evidence; v4 becomes the canonical producer on the real decision path. Behaviour-equivalent Python and TypeScript v4 verifiers ship against shared cross-language fixtures, so a customer’s stack verifies identically to ours. Tenant-admin custom semantic mapping packs (five framework packs, row-level-security isolated) and a Guardrails-AI-compatible validator adapter land as non-authoritative signals — never execution authority — with SBOM cross-check and DSSE artifact attestation over the validator supply chain.
Aug 21, 2026
Model Authority — Gates 0 through 6. A durable tenant-isolated model-authority registry with migration and admin API, a signed authority snapshot (eve.model.authority.snapshot.v1), versioned registry-entry digest canonicalization, execution-authorization token v2 with durable atomic single-use redemption under FORCE row-level security (concurrency-proven at-most-one winner), a provider-invocation recheck chokepoint with an adversarial test matrix, and tenant-isolated decision navigation over the durable execution ledger. Merged and deployed behind off-by-default flags; every gate fails closed.
Aug 21, 2026
Enterprise integration layer. /integrations rebuilt around the whole control point — Sources → Enforcement → Runtimes → Evidence — with Phase-2 adapters: four framework-only GRC normalizers and five governance-native runtime gateways. Behind it sits a canonical connector-readiness manifest (46 connectors across a six-class taxonomy, every entry grounded in a code reference) that CI enforces: a connector cannot be shown above its code-backed readiness, and a not-implemented connector cannot appear as available.
Aug 21, 2026
Vendor governance and evidence-return outbox. SOC 2 vendor governance and vendor change-management controls, plus a governance-bridge evidence-return outbox with reconciliation and provenance fail-close, so evidence handed to an enterprise GRC system can be reconciled rather than assumed delivered.
Aug 22–23, 2026
evecore.com becomes canonical. Brand consolidation onto evecore.com for search, with the brand redirect built but deliberately disarmed; documentation moves to docs.evecore.com; the SDKs default to api.evecore.com while api.eveaicore.com stays fully supported. A performance and security pass lands alongside: safe caching with correct Vary on compressed responses, fail-closed authorization and replay handling, and a batched billing drain.
Aug 25, 2026
Offline verifier proven; signing claims corrected. A standalone offline verifier demonstrates third-party verification with no call back to EVE, and stale signing claims are purged from public copy in favour of the KMS-resident signer actually in use. The owner-only billing control plane, master tier, and entitlement administration ship the same day.
Aug 26, 2026
Durable AI Inventory and GRC persistence — production activation. The AI System Inventory and the GRC system-of-record move onto PostgreSQL in production, paired with an RLS-safe runtime-role guard and a hard prohibition on runtime DDL in deployed environments. The inventory remains operator-declared by design: EVE does not scan or enumerate networks.
Aug 26 – Sep 1, 2026
Tenant-isolation program — measured, then armed. A tenant-table manifest with a structural posture guard and drift harness; an unbound-access probe to measure real exposure before flipping any policy; a verifier proving the serving task can never reach the schema owner; evidence-window instrumentation with traced nullable-tenant classification; a reversible per-table arm/disarm operator tool; and the first evidence-backed fail-closed batch — three tables now live under enforced row-level security, with the remainder sequenced behind the same evidence gate.
Aug 27, 2026
Governed lending authority. Fail-closed domain policy packs for lending decisions, plus a loan-origination-system adapter conformance gate — an adapter cannot ship without proving it honours the authority boundary. Off by default.
Aug 27–28, 2026
External guardrail engines, admitted as signals only. A governed guardrail-webhook adapter for agentgateway v1.4.x, and NVIDIA NeMo Guardrails 0.24 running inside the production image as an advisory signal that never holds execution authority. The architectural line is deliberate: third-party engines can inform a decision; CoreGuard alone makes it, deterministically.
Aug 30, 2026
Authorization-before-execution verified end to end. An authenticated agent proposing drop_database against a 48,219-row table returns BLOCKED citing deterministic rule identifiers — and the protected executor’s own invocation counter stays at zero. The same agent proposing query_data returns ALLOWED and the executor runs exactly once. The boundary is fail-closed: absent an ALLOW, nothing executes. Published as a reproducible demo and positioning article; S3 server access logging enabled across all buckets the same day.
Jul–Aug 2026
Governance security hardening campaign. An adversarial audit of the governance, CoreGuard, decision, sovereign, signing, and execution surfaces, run as finder → two-lens refuter with findings rejected by default. Every surviving finding was fixed fail-closed with regression tests, deployed, and smoke-checked — including a live decision-path required-fields fail-open, broken function-level authorization on 21 mutating router handlers, a cross-tenant authority-minting defect, webhook DNS-rebind SSRF, and an alg=none execution-binding forgery. The refuters dropped roughly 40 overclaims and forced one “fix” to be reverted in favour of the correct root cause. Full report published in the repository.
Sep 2026
Production signing custody moves inside the HSM boundary. Decision certificates, the unified audit bus, and published Merkle roots sign with ECDSA P-384 on an AWS KMS-resident key whose private half never leaves the KMS/HSM boundary. The self-describing signature tag keeps every previously issued certificate verifiable with no format break, and verification fetches the public key once and then runs locally — offline auditor verification makes zero calls to KMS, or to EVE.
Sep 1, 2026
Durable incident history on /status. Incident records persist as SOC 2 evidence rather than living in a dashboard’s memory, and Status is promoted to a standalone top-level navigation link — with availability reported at two explicit scopes (service-wide and per-component) so the two numbers cannot be quietly conflated.
Sep 1–4, 2026
Change management enforced in CI, not asserted in a policy document. An automated reviewer is wired in as a required approver, and a one-command change-ticket path ties every production change to an approved change record. The SOC 2 change-management control is therefore a gate that blocks a merge — not a written description of how changes are supposed to work.
Sep 5–6, 2026
Agent execution-surface coverage map. The operator-declared agent inventory is mapped against actual route reach derived from signed delegation grants, producing an honest coverage view: which execution paths CoreGuard governs, and which it does not. Declared by design — there is no scanner and no network enumeration, and the map says so on its face.
Sep 7–8, 2026
Endpoint Security — device trust as a governance input. Verified device risk from Microsoft Defender for Endpoint is bound into the decision context, so a compromised or non-compliant endpoint changes what its agent is authorized to do. Rollout follows the standard ladder — off → shadow → canary → enforce — with explicit, additive per-tenant enrollment. Shipped off by default.
Sep 9, 2026
Security Signals — govern on external findings without becoming a detector. External security findings arrive over an authenticated, tenant-bound ingest surface, are age-bounded, and are folded into authorization decisions. EVE deliberately does not become a detection engine: it governs on what verified detectors report, and a stale or unattributable signal cannot silently widen an agent’s authority. Shipped off by default.
Sep 10, 2026
Agent Authority Assessment — first productised engagement. A fixed-scope $2,500 assessment: an organization’s agent authority surface is analysed and delivered as a signed report generated on production infrastructure, signed by the KMS-resident key. The deliverable is independently verifiable rather than a PDF that asserts its own accuracy — which is the entire point.
Q3 2026 — Now
Enterprise pilot program. Sovereign SDK governance-as-a-service with first enterprise customers. Tenant-isolated Charter, Claims Ledger, Trust Dial, and Action Registry. Preparing additional provisional applications to extend coverage ahead of pilot.
Q3 2026
EVE Trust Services operational hardening. From controlled preview toward general availability: scheduled witnessing operations with monitoring and incident runbooks, auditor credentials and a published key-revocation feed, third-party security testing, and the contractual foundations for evidence custody.
Q3–Q4 2026
Specification drafting sprint. Converting identified coverage gaps into provisional filing packages across governance, integrity, forensic-replay, and trust modules as candidates are confirmed with patent counsel.
Q3 2026 — Now
Agent action firewall — enforcing on the authenticated action API; canary rung. Per-action execution authority, with a signed decision record per action. The scope is narrow and worth stating exactly, because three things get merged under this heading and only one of them blocks anything today. Enforcing: the authenticated /api/v1/agent-actions/* surface, which derives the tenant from verified authentication only and refuses an unauthenticated caller outright — for the one enrolled pilot tenant a disposition there genuinely stops execution, with a durable decision-record surface and certificates recorded at the firewall and payment-mint boundaries. Evaluating but not blocking: our own autonomous tool path, which is wired to the same firewall and records evidence, but presents an internal system identity that is deliberately outside the canary allow-list, so it proceeds regardless of disposition — functionally observe-only, and being closed under change control rather than by widening the allow-list, which would arm every autonomous action at once with no observation window. Not wired: the computer-use path, where authority is bound to verified screen state as a TOCTOU defence — a verified primitive, not a running computer-use firewall. Promotion between rungs is a deliberate, validated change rather than config drift: the flags live on the production task definition, so an infrastructure apply can never silently downgrade enforcement.
Q3–Q4 2026
Model-agnostic runtime — observation before enforcement. EVE governs, routes, and attests AI activity without depending on any vendor, model family, API format, or runtime. The governance core was already provider-blind; this closes the remaining gaps: a capability registry (EVE asks what capabilities a runtime exposes, never whether it is OpenAI or Anthropic — and an undeclared capability fails closed), a normalized error taxonomy so no vendor SDK exception ever reaches domain code, and a reference adapter implementing the full contract with no vendor SDK at all. Adding a provider is one adapter plus a capability descriptor: no change to CoreGuard, the proof chain, policy evaluation, or authority. An anti-lock-in test suite blocks every vendor SDK from the import path and proves governance still runs. The live-path ladder (off → shadow → compare → enforce) is merged and stays off in production; arming enforcement requires the mode flag and a separate explicit acknowledgement, and is gated on a compare-mode observation window.
Q4 2026 — 2027
FPGA/TPM proof-of-concept. The Hardware Veto Core running on programmable logic — sub-microsecond deterministic veto with tamper-evident attestation. The software side is done and is the actual prerequisite: the veto logic is already a pure, side-effect-free module with zero I/O and zero global state, and its firmware API contract is published as a C header. What exists today is the engagement scaffolding — a statement of work for the deterministic-veto bitstream, a contractor shortlist, and a published silicon-partner kit. No bitstream has been built yet; this is a funded-partner milestone, not shipped hardware.
Q4 2026
Attestation Marketplace launch. Multi-agent governance verification with Resilience Signatures, and the federated trust-exchange protocol live. The engine is implemented — roughly 2,500 lines across the governance and SDK layers, exposed through the trust-fabric API, with signed attestations, a revocation registry, and peer trust exchange behind a circuit breaker. Launch is gated on onboarding external attesters, not on writing the code: an attestation market with one participant is a log file.
Feb–Jul 2027
Non-Provisional Utility Filing. Full patent prosecution of the filed provisional applications and pending invention disclosures in omnibus conversion bundles. Conversion window: Feb 22 – July 2027. Micro Entity status throughout.
Q4 2026 – 2027
Open verification standard. The evidence vocabulary is already versioned and stable: 149 eve.* schema versions spanning decisions, delegation, authority reports, assurance, lineage, AI-BOM, workload identity, and compliance packages — with independent verifiers in Python, TypeScript, and Go that check a record against a published public key and nothing else. The next step is publishing the schema set and a conformance suite, so a third party can verify EVE-issued evidence and issue EVE-compatible evidence of their own with EVE nowhere in the loop. A governance standard that requires trusting its vendor is not a standard — the proof has to outlive the relationship, and the format has to be checkable by someone who has never spoken to us.
2027
Hardware enforcement research. Long-term research direction: compiling the deterministic veto core toward programmable logic and, potentially, purpose-built silicon. Exploratory — not on the near-term product roadmap.