The nonprofit Coalition for Health AI (CHAI) and the AIGovOps Foundation will share stewardship of OVERT (Observable Verification Evidence for Runtime Trust), an open technical specification developed by Glacis Technologies for verifiable operational evidence from AI systems.
CHAI explains that as AI systems move beyond retrieving, summarizing and analyzing information to supporting decisions and taking actions within healthcare workflows, organizations need more practical ways to understand how those systems operate.
OVERT is designed to complement existing governance policies, system logs and vendor documentation with standardized, independently verifiable evidence about AI actions captured within an implementation’s declared scope.
OVERT defines a common format for evidence generated when an AI system acts. It specifies how to create independently verifiable records showing whether declared governance policies, security controls and oversight requirements were applied at runtime, including the tool calls made by AI agents, without requiring protected content to leave the operator’s environment.
CHAI said it would work with its community to identify healthcare-specific needs and inform continued development.
Here is how this works: A healthcare organization requires that a given AI workflow have appropriate authorization, defined data boundaries and clinician review before it acts. With OVERT, runtime evidence documents whether those exact requirements were invoked for that specific AI action, creating a record that can later be reviewed by the organization or another authorized reviewer.
The AIGovOps Foundation is a nonprofit based in the Seattle area that develops open-source tools and a practitioner community for AI governance. Glacis is a company that develops runtime evidence infrastructure for AI systems. It is backed by AI House, Geoff Ralston’s Safe AI Fund (SAIF), Mighty Capital, Plug and Play and Lionheart Ventures
Once the stewardship arrangement formally takes effect, CHAI and AIGovOps will jointly oversee the OVERT specification. Glacis will initially maintain the specification and operate its registry.
“We wrote OVERT because we saw a proof gap in how critical AI safeguards are monitored, especially in healthcare,” said Joe Braidwood, founder and CEO of Glacis Technologies, in a statement. “We built it for everyone to use, our competitors included. That is why CHAI and AIGovOps will steward the standard, with Glacis as a non-voting editor. We will keep building implementations and compete on them like everyone else.”
“Healthcare organizations are asking for more technically rigorous ways to understand how AI operates within live workflows and whether defined safeguards are being applied,” said Brenton Hill, head of operations and general counsel at CHAI, in a statement. “By helping steward OVERT, CHAI can bring an open approach to decision-level evidence before our community for scrutiny, input and continued refinement. This is a tangible step toward giving organizations greater transparency into AI-enabled actions and greater confidence in how these systems are governed at scale.”
CHAI stressed that OVERT is a voluntary technical specification. It does not establish regulatory or legal requirements, certify the safety or effectiveness of an AI system, or determine whether an organization has complied with applicable law. Organizations remain responsible for defining appropriate controls and meeting their legal, regulatory and operational obligations. The collaboration does not establish a post-market monitoring network.
Leave a Reply