An article from IPWatchdog outlines a proposed International AI Assurance and Access Compact, a framework designed to let jurisdictions cooperate on AI verification without forcing developers to choose between oversight and intellectual property protection. The authors draw on precedents from the WTO, arms-control treaties, and environmental agreements, but stress that the Compact is a proposed arrangement, not existing law.
The operational tool would be a modular 'assurance passport': a verifiable record of findings tied to a specific model version, operator, configuration, and use. A receiving authority might recognize security testing performed abroad while separately examining local-language performance, discrimination risks, or remedies. Recognition would cover identified evidence, not every consequence of deployment, and material changes would trigger an audit of affected modules.
Confidentiality is treated as part of verification, not an afterthought. The article notes that U.S. trade-secret law and TRIPS Article 39(2) protect qualifying undisclosed information but do not create a blanket exemption from lawful regulatory examination. The proposed system would rely on restricted-access testing, limits on copying, purpose-specific permissions, and secure retention, while acknowledging that excessive secrecy could undermine meaningful scrutiny.
The article also cautions that safety testing does not clear training rights: a model can pass technical tests while its training materials remain subject to an IP dispute. A certificate would not convey a patent license or resolve a copyright dispute. The framework's usefulness would depend on whether accepted findings remain relevant to actual deployment, and it would introduce new costs around accreditation, equivalence, and regulatory capacity. The source presents this as a thought experiment rather than a settled policy proposal, and no other sources were available for comparison.