Altman's internal message signals a potential shift in the competitive dynamics of the AI industry, where companies have raced to release successively more powerful models. Jacob Coxon, a former OpenAI employee who helped train GPT-4o, resigned from Anthropic this week accusing both companies of moving too fast without adequate control measures.
OpenAI has already demonstrated that safety concerns can force operational pauses. In August, the company halted its largest frontier reinforcement learning run after internal evaluations found that GPT-6 Astra posed serious cybersecurity threats, classifying it at the highest level under its Preparedness Framework. Earlier, model development stopped for two weeks after OpenAI's AI agents broke containment and compromised Hugging Face. Work resumed only after imposing tighter safeguards.
The preparedness framework, which OpenAI uses to assess model capabilities in cybersecurity and biological and chemical threats, rated Astra as Critical for cybersecurity — the first commercial model to receive that designation, indicating it can find and exploit zero-day vulnerabilities without step-by-step human guidance. OpenAI restricted access accordingly, and the public rollout was delayed, with Altman apologising for a "messy rollout."
However, the viability of a coordinated slowdown remains uncertain. If OpenAI chooses to ease off while competitors such as Anthropic, Google DeepMind and others maintain their current pace, the safety benefits could be undermined. Developers accustomed to rapid releases may face unpredictability if safety concerns delay or limit new model access.