Independent third parties can now openly verify what autonomous AI agents are authorized to do and what they actually did, under a new open standard whose working draft went out for public comment on September 17, 2026. The Proof-of-Control v1.0 draft, built by Advanced AI Society alongside more than 80 global security leaders, will accept public comments through October 30, 2026 at advancedaisociety.org\/proof-of-control. Linux Foundation Decentralized Trust will host the standard as a lab, giving it a neutral home that no single vendor controls.<\/p>\n
Proof-of-Control sets three requirements for any AI agent that wants to claim it stayed inside its mandate. An agent’s authority has to be defined before it acts. The agent has to produce runtime evidence that it stayed inside that authority. And that evidence has to be auditable across the full lifecycle of the agent, not just at annual or quarterly check-ins.<\/p>\n
The design principle is that verification cannot belong to the verified. Unaffiliated, neutral third parties, as well as the developers and deployers of agents themselves, must be able to inspect the work agents do, rather than take an operator’s word that nothing went wrong.<\/p>\n
The standard builds on production-proven verifiable AI work from Advanced AI Society’s member companies. Co-design contributors included more than 80 security leaders drawn from across enterprise sectors.<\/p>\n
As those agents begin to move money, alter patient care, and touch critical infrastructure, an unchecked error can cascade through thousands of follow-on actions per second, multiplying the cost of every unverified decision.<\/p>\n
That gap between what agents do and what humans can inspect is what Advanced AI Society calls the Verifiability Gap. The alliance argues that traditional safety checks and audits, useful as they are, cannot keep up with agents that act in milliseconds. Trust in those agents requires continuous monitoring that any party can inspect, not periodic stamps of approval from a single vendor.<\/p>\n
Open verification applies open-source principles to the agentic stack, holding that components you cannot inspect cannot be secured and tasks you cannot verify cannot be delegated.<\/p>\n
Advanced AI Society has joined the Linux Foundation, and Proof-of-Control will sit inside LF Decentralized Trust as a lab. The cryptographic projects already housed there, including AnonCreds and Trust Over IP, supply the evidentiary backbone that the agent standard draws on.<\/p>\n
Housing the standard inside a neutral foundation is meant to prevent any one company from owning the referee. Cryptography’s core rule, that verification means showing the evidence rather than asserting the result, is meant to apply to autonomous AI as well.<\/p>\n
Distinguished Review Board members and senior advisors quoted in the announcement include cryptographer Bruce Schneier, former U.S. Commodity Futures Trading Commission Chairman J. Christopher Giancarlo, former U.S. Department of Homeland Security AI Policy Lead Noah Ringler, AstraZeneca Global Head of AI Cybersecurity Charles Iheagwara, and New York University Vice Provost for AI and Technology in Education Clay Shirky.<\/p>\n
Member and partner organizations supporting the work include Shielded Technologies, Oko Labs, Emme, and Solv Labs. Advanced AI Society’s co-founders are Tricia Wang (CEO), Michael Casey, and Sheila Warren, with Bettina Warburg as a co-founder. Ken Huang serves as Co-Chair of Proof-of-Control, and Jim Schwoebel directs research at the Open Verification Lab.<\/p>\n
The Proof-of-Control v1.0 working draft is open for public comment through October 30, 2026, with submissions accepted at advancedaisociety.org\/proof-of-control.<\/p>\n
Advanced AI Society and LF Decentralized Trust are co-hosting a public launch event on Wednesday, September 23, 2026, titled “Who’s Watching the Machines? Launching Open Verification, the Proof-of-Control Standard for the Agentic Era.” The event brings together alliance leaders, builders, deployers, and policy experts. Registration is available at advancedaisociety.org\/announcements\/webinar1.<\/p>\n
Proof-of-Control is an open standard requiring an AI agent’s authority to be defined before it acts, runtime evidence that it stayed inside that authority, and lifecycle-wide auditability of that evidence, so that any unaffiliated third party can verify the agent’s actions rather than rely on the operator’s word.<\/p>\n
Linux Foundation Decentralized Trust is hosting Proof-of-Control as a lab. The neutral foundation home is meant to keep open verification outside any single vendor’s control and to apply the same evidentiary rigor used in cryptographic projects such as AnonCreds and Trust Over IP.<\/p>\n
The Proof-of-Control v1.0 working draft is open for public comment through October 30, 2026 at advancedaisociety.org\/proof-of-control.<\/p>\n