kellerai.blogKELLERAI · AUTHOR
Jonathan Bowe
Chief Architect, KellerAI · AI Safety & Governance
Jonathan Bowe is Chief Architect of KellerAI, where he designs governed-inference systems for mission-critical organizations. His work centers on AI safety governance and evaluation frameworks: enforcing reasoning completeness through role-gated authority transitions, fail-closed degradation (“Boring Mode”), and replayable audit chains aligned to NIST AI RMF and DoD AI Ethical Principles.
Whitepaper Series
Published research on verifiable and auditable AI in production.
- Aviation and Banking Already Solved HallucinationThe AI field has been asking the wrong question. Aviation and banking each solved the underlying engineering problem decades ago — under regulatory compulsion, at enormous cost, and with a precision the AI industry has yet to borrow.
- What Changes When the Model ChangesA model upgrade is a controlled change, not a drop-in — and the vendor's benchmark is not your validation.
- The Audit You Can AuditWhy most codebase health checks change nothing — and what it takes to trust one.
- Trust but Verify: The Aviation Standard for Engineering with AIAviation didn't get safer by trusting pilots more — it built verification structures around them. AI-assisted engineering needs the same discipline.
- The Eval That Doesn't Follow the Model to ProductionYour eval is a measurement. Frontier models can recognize when they are being measured — and adjust their behavior accordingly.
- Why Self-Improving AI Needs a Trust DialAI systems are starting to improve themselves. Here is why that is dangerous, and what a fix looks like.
- The MCP Supply Chain You Forgot to GovernConnecting an agent to a public MCP server inherited a by-design execution surface the protocol's maintainer declined to patch.
Open Source
Specification-grade, policy-as-code governance infrastructure — written to be adopted and audited, not locked inside a product.
- opa-governance-libraryProduction OPA/Rego policy patterns for AI and multi-agent toolchains — circuit-breaker, audit-trail, and plugin-governance pillars, with a full test suite.
- agentic-telemetry-specA formal JSON Schema for telemetry across every Claude Code primitive — enabling provenance tracing, attribution, and backward-compatible decision logging.
- ai-provenance-specTamper-evident attestation for AI artifacts — a trusted release and deployment substrate for portable cognition capsules.
- trust-boundary-protocolA unified epistemic boundary layer that gates cross-domain agent actions on a shared view of every participant's confidence state.
- grounded-rag-specRepo-scoped, evidence-backed answering with explicit abstain and partial-answer contracts.
- kellerai-oss-templateAn OPA/Rego conformance authority and bootstrap scaffold for the kellerai open-source repository family.
- ood-math-handbookA curated mathematical-foundations corpus for out-of-distribution detection and certified-coverage problems in autonomous systems.
Background
Before KellerAI, Jonathan spent roughly twenty years flying C-17s, C-32As, the MD-88, and the Airbus A220 in high-consequence environments across the U.S. Air Force and commercial aviation. He helped build the AMC Digital Tactics Binder, adopted command-wide across Air Mobility Command. He served as Chief of Standardization and Evaluation, owning the AF Form 8 aircrew certification program for the USAF's only Prime Nuclear Airlift Force (PNAF) squadron. He also led a 210-person, 12-flight reorganization at the 1st Airlift Squadron, 89th Airlift Wing. That operating history underpins his central thesis: as base models commoditize, durable advantage accrues to organizations that enforce reasoning completeness in the decision-trace history and cited outputs a system accumulates over time.
Jonathan is based in Washington, DC, and welcomes conversations on AI governance, safety evaluation, and consulting engagements.