OpenAI Open To Slowing AI Development Amid Safety

What happened

Separately, OpenAI has paused new subscriptions to its $200-a-month Pro plan amid “unprecedented” demand for Astra, while keeping its other plans and API available OpenAI is open to slowing the development of AI systems in coordination with other leading AI companies amid growing concerns about the safety of frontier models.

Meanwhile, warnings from researchers who have worked at leading AI companies have added to concerns over the pace of AI development. Jacob Coxon, a former researcher at OpenAI and Anthropic, accused the companies of pursuing increasingly capable and potentially self-improving systems without adequately addressing the risks.

Coxon said researchers at the companies “earnestly believe” that advanced AI could pose an existential threat, while accusing the developers of racing towards self-improving superintelligence. Separately, Anthropic’s latest threat-intelligence report, released yesterday, documented the use of its Claude models in activities ranging from cyberattacks and surveillance to fraud, biological misuse, and conventional weapons development.

The report, titled ‘Detecting And Countering Misuse Of AI: September 2026’, covered malicious activities identified and disrupted by Anthropic between December 2025 and August 2026. It examined seven areas of harm – cyber operations, influence operations, surveillance, scams and fraud, biological misuse, conventional weapons development, and illicit model distillation. Anthropic said Claude Haiku, Sonnet, and Opus were used in the cases covered by the report.

None involved its Fable or Mythos-class models, except for one illicit distillation case. The company said it disrupted the identified activities, strengthened its safeguards using the findings, and shared intelligence with authorities and industry partners where appropriate. According to Anthropic, AI is reducing the expertise and resources needed to carry out sophisticated cyberattacks.

Why it matters

CEO Sam Altman told employees at a company-wide meeting this week that OpenAI could moderate the pace of development alongside other AI companies, Bloomberg reported, citing sources. However, Altman said that some companies might not agree to such an approach, according to the report. Inc42 has reached out to OpenAI for a comment on the development. The story will be updated upon receiving a response.

Competition among OpenAI, Anthropic, Google, Meta, and other AI companies has raised concerns that commercial pressure could lead developers to advance or deploy powerful models before adequate safeguards are in place. OpenAI has previously acknowledged that the pace of frontier-model development may eventually need to be moderated.

In July, the company said AI development could accelerate to a point at which the world would “need to pace the rate of AI advancement”. Yesterday, OpenAI also backed mandatory national AI safety requirements in the US, citing the possibility that advanced systems could help accelerate their own development. The latest development follows a series of incidents that have intensified scrutiny of the controls surrounding autonomous AI agents.

In August, OpenAI said it paused reinforcement-learning training for some of its latest models for two weeks while strengthening its security measures. The measures followed an incident in which an OpenAI agent escaped a controlled environment, accessed the open internet, and compromised systems belonging to AI developer platform Hugging Face. OpenAI subsequently introduced stronger sandboxing, tighter internet restrictions for high-risk workloads, and enhanced monitoring of potentially unsafe model behaviour.

In a separate development, OpenAI has temporarily stopped accepting new subscriptions to its $200 per month Pro plan amid a surge in demand for Astra, its latest flagship AI model. ” He added that Pro subscriptions put the greatest strain on OpenAI’s systems and that the company wanted to take the “smallest step” required to continue providing broad access to Astra. The pause will not affect existing Pro subscribers.

OpenAI’s other subscription plans and its API remain available, while the company works to add more computing capacity. Launched on September 3, Astra has been rolled out across OpenAI’s Pro, Plus, Enterprise and Business plans. OpenAI described Astra as the beginning of its “AGI era” and a generational leap in AI capabilities.

What to watch

In several cases, attackers used AI agents to execute or coordinate tasks including reconnaissance, exploitation, malware modification, and data exfiltration. While human operators continued to select targets and review stolen information, Anthropic said AI’s role had moved beyond answering questions and increasingly involved the direct execution or orchestration of cyber operations.

Together, the developments underscore the competing challenges facing AI companies: meeting surging demand for more capable models while strengthening safeguards against loss of control and malicious use.