OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei urged governments to adopt shared AI safety standards during a Wednesday session of the United Nations Security Council on September 23. They were in agreement: The AI industry needs international standards and cooperation, or risk global security. CNN reported that Amodei warned that if not, AI could be a risk to humanity as a whole. Altman said rapid model progress has made the timeline feel more compressed, the upside more tangible, but also the stakes and the risks more immediate. One day later, The Information reported that Google, OpenAI and Anthropic have agreed to stand up a FINRA-style self-regulator—tentatively the Frontier AI Standards Agency—with no government oversight, targeting a launch by end of 2026 or early 2027.
I do not run a self-regulatory organization. I run a construction ERP at coenconstruction.com, an estimating tool at estimate.pro, and a permit-drawing platform that went live five days ago. The ERP calls OpenAI for invoice parsing. The estimator calls Anthropic Claude for blueprint line-item extraction. The permit tool calls Google Gemini for OCR on hand-drawn site plans. All three integrations depend on API endpoints I do not control, running under terms the vendor can rewrite, inside models the vendor can deprecate when the product roadmap shifts. If the three vendors that supply every API I call decide to write their own safety standards, audit their own models, and staff their own testing body, I do not get a vote. I get the standards they publish and the audit results they choose to share.
The candidate said there will be no FDA for AI
Google, OpenAI and Anthropic have approached Sriram Krishnan to serve as chief executive of what is tentatively called the Frontier AI Standards Agency. Krishnan left the White House AI adviser role in June 2026 after publicly rejecting the idea of a licensing regime, saying there will not be an FDA for AI. On the way out he said a centralized agency requiring a team of lawyers before you can get a model out would put sand in the gears of the AI revolution. AI Weekly noted the three biggest frontier labs want to build their own regulator, and they want the job run by a man who spent his time in office arguing Washington should not build one.
The construction ERP has an incident-reporting workflow. When an invoice extraction fails, the system logs the failure, flags the API call, and emails me the raw response. The workflow exists because I need to know when the integration breaks, not because OpenAI asked me to report it. If the Frontier AI Standards Agency publishes an incident-reporting standard, I will read it. If the standard says I must report every hallucinated line item or every refusal to parse a scanned receipt, I will evaluate whether compliance costs more than switching vendors or building the extraction logic in-house. The agency will not force me to comply because the agency is not a regulator. It is a private body through which American brokers police themselves, under the oversight of the SEC, except the SEC in this analogy is three AI labs and the oversight is voluntary.
The funding comes from the companies being audited
The three AI companies are working to create an organization in which builders of frontier AI models would regulate themselves, without government oversight. The standards body would support third-party organizations that test models before they are deployed, would detail how AI developers should report safety and security incidents, would define the labs' voluntary safety and security commitments, and would establish qualifications for independent auditors of models and labs. Funding would need to be substantial and likely mostly come from industry to attract top talent and gain access to the compute it needs. The proposal is modeled on FINRA, the securities industry self-regulator that levies fines, suspends licenses, and reports to the SEC. The Frontier AI Standards Agency will have no SEC. It will have shared technical evaluations, pre-release audits, independent testing frameworks and standardized safety protocols, all funded by the labs being audited and enforced through standards the labs agree to follow.
The permit platform makes six thousand Gemini API calls per month. The calls cost $0.04 each. If Google decides that funding a self-regulatory safety body requires raising API prices by ten percent, my monthly bill goes from $240 to $264. If Anthropic and OpenAI raise their prices by the same margin to fund the same body, the construction ERP and the estimating platform each pay the increase. I do not know whether the Frontier AI Standards Agency will assess its funding as a percentage of API revenue, a flat annual membership fee, or a per-audit charge. I do know that the cost of the audits will not come out of the labs' profit margins. It will come out of the budget of every developer calling the APIs being audited. I wrote on September 17 about the legal boundaries of coordination between competitors. This is not coordination on price. It is coordination on the cost structure underneath the price, which the customer pays either way.
The three labs asked the UN for international oversight on Wednesday. On Thursday they announced they would regulate themselves instead, with no government role and funding from the companies being tested.
Cohere CEO Aidan Gomez offered a pointed critique, warning on X that letting the biggest AI companies jointly set rules risked putting smaller competitors at a disadvantage. They initially aimed to form a public-private partnership, but that effort stalled in the Trump administration. The alternative is a self-regulator with no statutory authority, no enforcement power beyond voluntary compliance, and no funding source except the three labs that will write the standards they audit themselves against. The construction ERP will keep calling OpenAI. The estimating tool will keep calling Claude. The permit platform will keep calling Gemini. The APIs will keep working until they do not, under terms that can change when the vendor and the self-regulator it funds both agree the standard should be updated. I will learn about the update when the API returns a new error code or the monthly bill includes a line item for regulatory compliance.