OpenAI is talking with multiple independent assessors to help grow the number of companies that have expertise in frontier AI safety and can provide effective third-party assessments, the company said in a blog post published Tuesday, September 22, as reported by PYMNTS.

The post lays out OpenAI's proposed priorities and principles for effective third-party oversight — a framework it says should apply to all frontier AI labs as they train, evaluate and deploy increasingly capable models. The announcement lands in the middle of an unusually charged month for AI safety policy, and it reads as both a concession to critics and an attempt to shape the rules before governments write them. For anyone following breaking AI news, it is the latest move in a weeks-long contest between the labs, their critics and regulators over who gets to verify that frontier models are safe.

What OpenAI Is Proposing

"We are committed to supporting independent assessors and establishing clearer, shared international standards — both through future laws and private governance initiatives — for effective third-party assessments," OpenAI said in the post, according to PYMNTS.

The core problem OpenAI describes is a market failure: there are simply too few organizations with the technical depth to audit a frontier model. A meaningful safety assessment requires access to unreleased systems, expertise in areas from cybersecurity to biosecurity, and the institutional independence to publish unwelcome findings. By recruiting and supporting multiple assessors rather than a single privileged auditor, OpenAI says it wants to build an ecosystem where third-party evaluation is a routine part of deploying frontier systems — not a one-off public relations exercise.

The company did not name the assessors it is talking with, and it remains to be seen whether the resulting arrangements will give evaluators the access, funding independence and publication rights that safety researchers consider meaningful.

A Week of Whiplash on AI Policy

The proposal arrives amid a rapid sequence of policy developments that have defined September's AI debate:

  • Monday, September 21: OpenAI called for the United States to lead an international effort to develop global technical standards for frontier AI — including recursive self-improvement, the point at which AI systems improve themselves. The company warned that fully autonomous AI development could outpace humans' ability to understand and control it, and said decisions about whether and how to proceed need to be made now.
  • Friday, September 18: California Governor Gavin Newsom signed an executive order convening national experts to reinforce the state's AI safety laws, explicitly aimed at accelerating implementation of new third-party oversight of safety and security risks in AI systems.
  • Saturday, September 19: President Donald Trump rejected the industry's safety concerns in a Truth Social post, writing: "We will not in any way hinder or stifle the Growth of this incredible industry. AI is the next Industrial Revolution, or Internet, but will be even larger and more impactful, possibly as much as 25% of our Country's GDP."

As we reported earlier this week, OpenAI's US-led standards proposal marked a striking shift for a company that spent years resisting binding oversight — and the new third-party assessment principles extend that posture onto the private governance front.

The Independence Problem

OpenAI's embrace of external watchdogs follows a parallel proposal from its chief rival. About a week earlier, Anthropic CEO Dario Amodei published an essay proposing an AI safety plan built around independent evaluators embedded inside leading AI companies, coordination among companies in democratic countries, and eventually an international agreement that includes China.

Both proposals have drawn immediate skepticism from experts. CNBC reported that a coalition of more than 100 AI experts is urging independence and transparency from Anthropic, OpenAI and other foundation model labs conducting evaluations, warning that evaluators who are embedded within — or paid by — the companies they audit face unavoidable conflicts of interest. Watchdogs backed by AI pioneer Geoffrey Hinton have pressed both companies on the details of their safety plans, and commentators have flagged the revolving door between AI safety evaluation organizations and the labs themselves: the people grading the safety homework may also be the ones who wrote it.

That tension is the crux of the debate OpenAI's new post tries to defuse. The company's answer — more assessors, shared international standards, and oversight enforced through both future laws and private initiatives — is a framework, not yet a guarantee. Critics will point out that the labs proposing the rules are the same labs subject to them, and that self-shaped governance has historically favored the companies doing the shaping.

Why It Matters

The next frontier models are already in training, and the window for designing oversight before deployment is closing. If OpenAI's recruitment of independent assessors produces auditors with real access and real independence, it could become the template for how frontier AI is governed worldwide — a private regime that lawmakers in Washington, Brussels and elsewhere have so far failed to build themselves. If it produces affiliated consultancies with NDAs, it will confirm the skeptics' worst reading.

Either way, the fact that OpenAI is competing with Anthropic over who can propose the most credible oversight regime is itself new. A year ago, the labs were unified mainly in opposing regulation. Now they are racing to define it.

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →