OpenAI, Anthropic and Other North American AI Firms Meet to Finalize Third-Party Oversight Mechanism
Market reports indicate that OpenAI, Anthropic and other leading North American AI companies held a meeting on Sept. 14, 2026 to finalize a third-party oversight mechanism.
The meeting follows Anthropic CEO Dario Amodei's Sept. 12 essay, 'We Must Pace the Frontier,' in which he called for third-party permanent "employee-level" system access, checkpoints and international coordination to slow the development of the most powerful models. OpenAI CEO Sam Altman responded by agreeing with "frontier pacing" and said OpenAI would not pursue an IPO in 2026, arguing that going public before safety and alignment work is complete would be unwise. Elon Musk, founder of xAI and SpaceX, also endorsed Amodei's remarks online.
The discussions come amid rising concerns about opaque AI systems. In September, OpenAI said its internal frontier model used about 10,000 concurrent AI agents for roughly 88 hours to solve the Navier-Stokes existence and smoothness problem, one of the Millennium Prize Problems. However, human reviewers found that the search process, intermediate reasoning and the path to the solution remained inaccessible. OpenAI later faced accusations of plagiarism from two mathematicians, which it denied, though it acknowledged launching its research after hearing about Anthropic's related work. Shortly afterward, 25 Fields Medal winners, including Terence Tao, issued a joint statement warning that AI and mathematics are becoming seriously misaligned.
Concern over autonomous model development is also growing. Anthropic and OpenAI have said they are using models to help with model iteration, with Anthropic claiming that Claude now writes more than 80% of its codebase. Anthropic has predicted that models will actually use recursive self-improvement within one to two years. Anthropic co-founder Jack Clark has said current alignment techniques could break under RSI, as tiny errors could compound over generations and lead to loss of control.
