Anthropic, the company behind the Claude AI chatbot, has acknowledged that it has crossed a major artificial intelligence safety threshold just days before its planned stock market debut. The company disclosed the development in an amended IPO prospectus filed with US regulators on November 10, raising fresh questions about the pace of AI development and the ability of humans to maintain control over increasingly advanced systems.
The safety threshold is part of Anthropic’s Responsible Scaling Policy, which requires additional safeguards as its AI capabilities advance. The company said it had determined that the threshold had been met based on its expectations for future models and the overall pace of technological progress, rather than solely on the capabilities of current systems.
AI Is Accelerating Its Own Development
The triggered threshold relates to AI’s growing contribution to Anthropic’s research and development activities.
Under the company’s safety framework, the threshold is reached when AI contributes enough to its own development to approximately double the speed at which its models improve. This would effectively compress two years of AI progress, based on the 2018-2024 growth rate, into a single year.
Anthropic acknowledged that AI tools are already accelerating its research and development processes.
An independent nonprofit research organisation, Model Evaluation and Threat Research (METR), had also reported that Anthropic’s AI development capabilities were improving nearly twice as quickly as they were in October 2025.
The organisation estimated that much of the acceleration could be attributed to AI systems helping researchers develop more advanced models.
Anthropic Says Human Oversight Remains Essential
The development raises concerns about whether AI systems could eventually improve themselves faster than humans can evaluate their behaviour and potential risks.
Anthropic said it does not currently expect to safely use Claude to autonomously design and build successive generations of advanced AI models without meaningful human supervision or understanding.
The company is also considering changes to its safety policies to ensure humans retain control over critical research and development decisions.
In its IPO filing, Anthropic warned that increasing reliance on Claude could reduce the organisation’s visibility into certain internal processes. It cautioned that some activities could eventually become difficult for individuals or groups of humans to understand and control.
Crossing the safety threshold does not mean Anthropic will stop developing AI models. Instead, it signals the need for additional safeguards and monitoring as AI systems take on a larger role in research.
Independent Researchers Question Earlier Safety Assessment
METR had previously reviewed Anthropic’s assessment of its AI development capabilities and concluded that the company’s August report may have understated the extent of acceleration.
The independent organisation said it was plausible that the safety threshold had already been crossed by mid-July.
Anthropic has acknowledged the importance of external evaluation but has also limited the disclosure of certain details, citing commercial sensitivity and the possibility of providing competitors with information that could accelerate their own development.
$1.4 Trillion IPO Adds Investor Attention
The safety disclosure comes as Anthropic prepares for a potentially record-setting IPO.
The company has set an indicative share price range of $70 to $78, implying a valuation of approximately $1.4 trillion when all potential shares are considered. Trading could begin on Nasdaq as early as November 19 under the ticker symbol ANT.
Major technology
