Joe Braidwood
Joe Braidwood is Founder & CEO at Glacis Technologies, Inc., where he leads the development of AI supervision systems for healthcare and maintains the OVERT open standard alongside industry partners. He previously founded Yara AI, Vektor AI, Scener Streaming Service, and Spider Design Limited, and served as Chief Strategy Officer at Vektor Medical and Chief Marketing Officer at SwiftKey, which was acquired by Microsoft. His work has focused on building AI products and advancing accountability in AI.
Glacis Technologies Marks One Year, Launches Auditable "Flight Recorder" for AI; Pre-seed Closed and Lloyd's Lab Backing
Glacis Technologies, now one year old, has built an auditable "flight recorder" for AI that records what systems did and which safeguards ran so skeptics can verify behavior without trusting vendors. The company closed a pre-seed, joined Lloyd's Lab cohort 17, contributed the OVERT open standard co-stewarded by CHAI and AIGovOps, engaged with regulators and health systems, and is moving toward its first clinical production workflow.
Joe Braidwood announces Utah’s AI sandbox requires verifiable signed receipts using GLACIS’s OVERT standard
Utah now requires AI systems in its sandbox to produce verifiable, signed receipts proving safety controls ran, using the OVERT evidence standard created by GLACIS Technologies, marking a first-of-its-kind regulatory evidence requirement for AI risk management.
Utah requires Glacis-backed cryptographically signed receipts (OVERT) for every covered clinical AI inference in its AI Learning Laboratory
Utah’s Office of Artificial Intelligence Policy will require designated vendors in its AI Learning Laboratory to issue cryptographically signed receipts—based on the OVERT standard—for every covered clinical AI inference or control evaluation, as a condition of participation in the sandbox.
Glacis hands OVERT governance to CHAI and AIGovOps; Glacis becomes non-voting editor
Glacis has transferred governance of the OVERT standard to the Coalition for Health AI (CHAI) and the AIGovOps Foundation, becoming a non-voting editor to remove its commercial control. The agreement makes CHAI and AIGovOps Co-Stewards with equal voting power over releases, normative proposals, appointments, and budgets, while Glacis retains editorial participation but no vote or authority over conformance criteria.
Glacis Joins Lloyd’s Lab Accelerator Cohort 17 After Pitch Day Selection from 280 Applicants
Glacis was selected as one of ten startups for Lloyd’s Lab Accelerator Cohort 17, chosen from a record 280 applications, and will spend ten weeks starting 28 September working with market participants on an AI-related insurance product; photos and a short note from Pitch Day accompany the announcement.
Glacis joins PACT AI as founding member to advance independent assurance for consequential AI
Glacis joined PACT AI as a founding member to help build independent, credible assurance for AI systems that take real-world actions, contributing to governance, professionalization, and a market for recognized assurance.
Joe Braidwood: Use Signed, Verifiable Runtime Records to Strengthen AI Security Controls
Joe Braidwood argues that AI security should pair detection and dashboards with signed, verifiable runtime records so outsiders (regulators, auditors, counsel) can independently check integrity, scope, and attribution of reported control outcomes.
Joe Braidwood: AI Governance Sets Intent; Operational Supervision Provides Proof Through Scoped Evidence
Joe Braidwood explains that AI governance sets the policies, roles, processes, and controls that define how AI should behave, while operational supervision provides the scoped evidence and records showing those controls actually reported on covered actions.
Joe Braidwood: Anthropic’s Fable 5 Suspension and Restoration Show Need for Scoped, Verifiable Operational Evidence
Joe Braidwood uses Anthropic’s brief June–July 2026 suspension and rapid restoration of Fable 5 to argue for the need for scoped, independently checkable operational evidence (control telemetry, artifacts, and reproducible tests) to evaluate contested AI incidents and safeguard effectiveness.
Joe Braidwood: AI Governance Maturity Means Event-Level Proof, Not Just Written Policy
Joe Braidwood argues that true AI governance maturity requires more than documented policies—it needs independently checkable, event-level evidence showing configured controls, decisions, and actions in live systems. Relying on signed records and artifacts inflates self-assessments because those documents don’t prove coverage, control effectiveness, safety, or compliance.
Joe Braidwood: Use Scoped Signed Records to Reconstruct Prompt Injection Incidents After Anthropic Fable 5 Safeguard Bypass
Joe Braidwood argues that because prompt injection is always possible, defenders should maintain scoped, signed records showing which controls allowed, held, or denied actions and where coverage gaps exist so incidents can be reconstructed without confusing a signature for proof of effectiveness.
Joe Braidwood: AI Governance Faces a 'Verification Gap' — Policies Lack Checkable Evidence of Controls
The article argues that beyond policy, AI governance faces an operational "verification gap": teams often lack checkable, scoped evidence tying intended rules to what configured controls actually reported. It outlines common governance obstacles—pace, scope, and stakeholder alignment—and stresses the need to connect rules to verifiable operational evidence.
Joe Braidwood: Signed AI Attestations Verify Records but Don’t Prove Safety, Coverage, or Signer Independence
Joe Braidwood explains that signed AI attestation records let third parties verify specific reported facts (like unchanged bytes, matching signatures, and reconstructed event sequences) but cannot by themselves prove control effectiveness, coverage, safety, data locality, or signer independence.
Glacis Links AI Governance Intent to Operational Evidence but Stops Short of Proving Safety or Compliance
AI governance platforms catalog policies, models, owners, and approvals but stop short of proving how controls actually behaved in real operations; Glacis links governance intent to scoped operational evidence for specific AI actions without claiming to prove safety or full compliance.
Joe Braidwood: Signed OVERT Records Enable Verifiable Runtime-Control Results with Bounded Disclosure
Joe Braidwood explains that a signed OVERT record provides tamper-evident, bounded evidence (hashes, outcomes, signatures, metadata) so third parties can verify reported runtime-control results without disclosing protected payloads, while noting the record doesn't prove completeness or effectiveness and payload locality depends on deployment.
Glacis proposes signed-record operating boundary to secure AI agents and link controls to actions
The author argues for an operating boundary for AI agents that lets routine actions proceed while holding, narrowing, or denying risky exceptions and producing signed, verifiable records tying controls to agent actions so organizations can audit what actually happened.
Joe Braidwood: AI Governance Documentation Alone Doesn’t Constitute Operational Evidence — Preserve Signed Runtime Records
Joe Braidwood argues that governance documents (policies, narratives, logs) show intent and process but are not the same as operational evidence; organizations should preserve signed, scoped runtime records that provide checkable integrity, attribution, and event-specific outcomes alongside documentation.
Joe Braidwood: Signed Operational Records Can Provide Verifiable Evidence for AI Security Controls but Don’t Prove Full Coverage
Joe Braidwood argues that AI-driven security controls lack an independent, verifiable evidence layer: signed operational records could make bounded claims about specific control events checkable, but they don’t prove full coverage, effectiveness, or signer independence.
Joe Braidwood proposes standards-based framework for underwriting healthcare AI risk with illustrative composite cases
Joe Braidwood presents a standards-based research framework to assess and underwrite risks from healthcare AI, illustrated with composite cases and not representing an insurance commitment.
Joe Braidwood: Glacis Builds Infrastructure to Ship Safer AI, Updates OVERT to Version 1.1
Joe Braidwood recounts Glacis’s development of auto-redteam and OVERT—now updated to OVERT 1.1—and explains how the team built infrastructure to enable safer, supervised AI with bounded operational evidence.
Joe Braidwood: Anthropic Rewrites Responsible Scaling Policy; Voluntary AI Safety Rules Remain Fragile
Joe Braidwood argues that Anthropic revised rather than dropped its Responsible Scaling Policy, highlighting that voluntary, self-policed AI safety commitments are unstable and should be complemented by operational evidence.
Joe Braidwood’s GLACIS Introduces Signed Operational Records to Prove Healthcare AI Behavior at ViVE 2026
Joe Braidwood highlights that at ViVE 2026 healthcare AI moved from novelty to scrutiny, with buyers demanding verifiable operational records showing what models did. He presents GLACIS as a system that links runtime controls to signed, checkable records (hashes, signatures, outcomes) so authorized reviewers can verify covered events while keeping protected payloads inside operators’ environments.
SeaHealthTech Panel Says Hospitals Need Evidence Not Promises as AI Adoption Surges in Healthcare
A Seattle panel of healthcare leaders argued that as AI adoption accelerates across hospitals and Medicare plans, organizations need independently checkable evidence tied to specific decisions rather than vendor promises to ensure safety, accountability, and compliance.
Joe Braidwood: JPM 2026 Signals Demand for Documented Healthcare AI Supervision as Colorado, EU Laws Tighten and Litigation Rises
Joe Braidwood highlights a shift at JPM 2026 from asking whether organizations use AI to demanding documented, supervised evidence of how AI shapes healthcare decisions, noting new laws (Colorado SB 26-189, EU AI Omnibus) and rising litigation risks like the Sharp HealthCare ambient-scribe case.
Joe Braidwood on GLACIS: Evaluating Whether ISO 42001 AI Management System Certification Is Worthwhile for Vendors
Joe Braidwood evaluates the costs, benefits, and limitations of obtaining ISO 42001 certification for AI vendors, providing a balanced analysis to help organizations decide whether certification is worthwhile.
Organizations
See all 8-
Glacis Technologies, Inc. Founder & CEO 2025–present · current
-
Yara AI Co-founder & CEO 2024–2025 · past
-
Vektor Medical Chief Strategy Officer 2021–2024, Co-founder, Vektor AI 2021–2024 · past
Education
-
Brown University Master, Public Affairs, Public Policy · 2007–2008
-
University of Cambridge BA, MA, Law · 2004–2007