Aevion.ai

Aevion

Proof-native governance for autonomous AI systems.
_{Machine-checked invariants. Cryptographic receipts. Byzantine-resilient consensus.}

Aevion builds the governance layer beneath AI products ΓÇö the formal, receipted, independently auditable infrastructure that sits between an agent's decision and the world it acts on. Not guardrails. Not classifiers. A proof-native control plane.

On June 9, 2026, NIST published a mathematical proof (Vassilev, IEEE S&P) that no finite guardrail set can be universally robust against adversarial AI. NIST's official guidance: transition to a continuous-monitor-and-update security model. The same day, Anthropic launched Fable 5 / Mythos 5 ΓÇö same model, different access envelopes ΓÇö and publicly stated universal jailbreak prevention is "likely impossible."

Aevion's architecture was already built to this specification: receipt chain (continuous monitoring), counsel colony (proactive red-teaming), halt gate + human escalation (operational resilience). We do not claim to have defeated the impossibility theorem. We claim to have built the architecture the theorem says you need ΓÇö and published the exact list of what remains unproven.

Scott Leishman ΓÇö Founder & Principal Investigator

U.S. Navy Air Traffic Controller (2003ΓÇô2014), combat-zone deployment to Camp Lemonnier, Djibouti. M.S. Aeronautics, Embry-Riddle Aeronautical University (2018). B.S. Applied Science and Technology, Thomas Edison State University. B.S. Business Administration - Finance, Southern New Hampshire University. B.S. in progress ΓÇö Information Technology (Cybersecurity), Arizona State University.

Research. "Air Traffic Control Human Factors with Drones" — M.S. Capstone, ERAU (2018), peer-reviewed for final grade. SHEL model + ANOVA quantitative analysis against NASA Ames simulation data. Tested the hypothesis that UAS incorporation significantly impacts controller performance. Covered workload measurement, NextGen technologies (ADS-B, UTM, LATAS, LAANC), and loss-of-separation risk thresholds under BLOS automation — direct intellectual precursor to Aevion's Koopman ρ threshold derivation for autonomous system drift detection.

Additional graduate research (2015ΓÇô2019): ScanEagle risk assessment (MIL-STD-8820), BLOS operations and SATCOM human factors, GCS automation and skill retention, detect-and-avoid separation technologies. Full archive: unmannedac.blogspot.com.

Operational aviation safety discipline ΓÇö where traceability, command authority, and zero-failure operating discipline are mission-critical ΓÇö now formalized as machine-checked proof obligations for autonomous AI systems.

Aevion-Verifiable-AI is the canonical monorepo for the whole company. All other Aevion-ai repositories are private satellites, worktrees, or historical branches of that single source of truth. The only public-facing artifact is ProofOS, a governed mirror of an allow-listed subset pulled from the monorepo.

All other repositories under Aevion-ai are private and are not intended for public use. See Aevion-Verifiable-AI/docs/ops/repository_strategy.md for the full inventory and mirror policy.

Aevion LLC ΓÇö Service-Disabled Veteran-Owned Small Business
CAGE: 15NV7 ┬╖ UEI: JFCXAGHB3QM6 ┬╖ St. Cloud, MN
NIST AI Consortium applicant ┬╖ SBIR/STTR eligible
Contact: scott@aevion.ai ┬╖ aevion.ai

Theorem where provable. Assumption where modeled. Hypothesis where empirical.
Receipt where observed. Gate where safety-critical.