About Glacis

The controls already exist. The evidence doesn’t.

AI systems are already drafting the clinical note, screening the candidate, triaging the prior authorization. Guardrails run, policies are written, review boards meet — and still nobody can prove what happened at the moment of the decision. Glacis exists to close that gap.

Start free Talk to us →

Why we exist

It starts with a product we shut down.

Our founder built a generative-AI mental-health product because he believed language models could reach people the mental-health system cannot. He shut it down with an investor ready to write a check — not because he doubted the safeguards he had built, but because he had nothing to show anyone.

He wrote about that decision publicly, and it traveled further than he expected. Regulators from the UK, France, Australia and several US states got in touch to ask how the rules ought to be written. Glacis is the answer he owed them. Within a year of incorporating, a state AI office, a Lloyd’s-market carrier, a national standards coalition and regulators from several jurisdictions had all come to us independently.

The conviction underneath the company is a simple one. The controls already exist: guardrails run, policies are written, review boards meet. What does not exist is evidence — a record of what a system actually did at the moment it acted, that somebody outside the company can check for themselves. Until that record exists, every assurance about a high-stakes AI system is a claim made by the party with the most to lose, and duty of care is something you assert rather than something you can show.

This matters because the fallback is to slow the technology down or ban it outright. Accountability infrastructure is what lets useful clinical AI be deployed at all. That is the whole argument, and it is why we think an evidence substrate underneath high-stakes AI is inevitable rather than optional.

“We’re all highly aligned on the inevitability of this evidence substrate in high-stakes AI systems.”

Joe Braidwood, Co-Founder and CEO

Today the platform runs in production inside customer infrastructure, minting runtime receipts, and we are working with the Digital Medicine Society on its Operationalizing AI Governance project. Governance sets the standard; independent evidence proves it was met. How that works in practice — controls that sit inline at the inference boundary, the runtime receipt a governed action emits, the evidence pack those receipts roll into — is documented on the product pages. This page is about why we built it.

In numbers

Numbers we can stand behind.

50+Combined years in AI
OVERTStandard authored, not owned
1M+Records signed
ZeroCustomer data we ever see

Leadership

Three founders who have had to show their work.

A founder who shut down his own AI product rather than overclaim, a practicing child psychiatrist who has sat inside a first-of-its-kind FDA review, and an engineering leader who spent nearly two decades building the cloud infrastructure everyone else now stands on.

Joe Braidwood

Co-Founder & CEO

On the founding team at SwiftKey, one of the earliest natural-language-modelling companies, whose predictive keyboard reached hundreds of millions of devices before Microsoft acquired it and passed a billion afterwards. Head of strategy at Vektor Medical, where he incubated the AI team and led the reimbursement effort that secured a CPT code for an AI-based cardiac mapping product. Then he built a generative-AI mental-health product and shut it down on his own judgment when he could not prove its safeguards held. He led the authorship of OVERT v1.1, the open runtime-evidence standard this company deliberately does not own. Close to seventeen years of language-model products, and a Cambridge law background behind the way we reason about duty of care and admissible evidence.

Language Models Since 2009 Regulated Healthcare Cambridge Law

Rohit Tatachar

Co-Founder & CTO

Nearly two decades at Microsoft, most recently on the Azure AI Foundry team, where he owned both the engineering and the product side of his portfolio and led services that grew from tens of millions to more than a billion dollars in revenue. He left because he had watched, from the infrastructure side, exactly where the gaps were opening in what the hyperscalers were shipping, and he wanted to bring the safety side to a worldwide audience. Architect of the Glacis runtime kernel, which is cloud-agnostic and model-agnostic on purpose: a proof layer that is not independent is not proof.

Hyperscale Infrastructure Azure AI Foundry Runtime Kernel

Jennifer Shannon, MD

Co-Founder & Chief Medical Officer

A physician and child psychiatrist with more than twenty years in clinical practice, who still sees patients part-time to keep herself real. She trained in psychiatry at the University of Washington, completed her child psychiatry residency and fellowship at Seattle Children’s, and remains courtesy teaching faculty at the UW School of Medicine. She served as a medical director at Cognoa, where the team earned FDA De Novo authorization for Canvas Dx, an AI-based autism diagnostic, and she has since contributed monitoring, baseline-behavior and ownership language into the Coalition for Health AI’s agentic AI best-practice framework. She knows what a governance committee actually asks, and where the clinical evidence bar sits, because she has stood on the other side of it.

20+ Years in Practice Clinical AI Safety Health AI Standards

Advisors

People who have done it before.

Operators, researchers, and journalists who have built trusted institutions and know what it costs to earn that trust.

Anil Karmel

Co-founder, RegScale

Co-founded RegScale, the AI-powered cybersecurity GRC platform, after co-founding and exiting C2 Labs. Invented Los Alamos National Laboratory’s Infrastructure on Demand.

Selvan Senthivel

Chief Technologist, GE HealthCare

Twenty years architecting production AI at scale, leading engineering organizations of 400+. Previously led AI/ML engineering at AWS, including Comprehend Medical.

Dan Preston

CEO, Stand Insurance

Founder and former CEO of Metromile, which he took public. Now building Stand Insurance. Knows what it takes to underwrite a category nobody has priced before.

Dávid Márton

Head of Data & AI, Atria Health

Leads data and AI at the Atria Health and Research Institute in New York, after computational neuroscience and AI research at Harvard and clinical AI at Vektor Medical.

John Ryley

Former Head of Sky News

Ran Sky News for seventeen years. Knows how institutions earn public trust, and how quickly they lose it.

Sri Chandrasekar

Managing Director, AI House

Nearly a decade at Point72 Ventures, where he helped build the ventures and private equity businesses, after leading investments at In‑Q‑Tel for the intelligence community. He wrote the first check into AI House, where Glacis started.

What we stand for

Four things we will not trade away.

Proof, not promises

Trust means believe me. Proof means we have shown you. When an AI system touches patient care, “trust us” is not a governance model.

We refuse the overclaim

A runtime receipt proves that specific claims were made about a specific event, checkable independently. It does not prove a control is effective or a system safe, and a vendor who says otherwise is the problem we exist to fix.

The standard isn’t ours to keep

We wrote OVERT and then put its stewardship outside the company, so the standard can outlast any single vendor, this one included.

Your data never moves

Controls run inside your environment and classifiers run on your own hardware. The evidence belongs to your team, and anyone you hand it to can verify it without asking us.

The standard

Built to outlast us.

We wrote the OVERT standard for runtime evidence, then put its stewardship outside the company so it can outlast us. If you’re working out where you stand with AI, we’d like to hear about it.

Book a 30-minute assessment

Backed by