Ahead of IPO, Anthropic says safety-monitoring demands pushed customers to other providers

Anthropic said safety-monitoring and data-retention rules for some advanced models have caused customers to move to other providers or deployments where the company has less visibility, showing a commercial cost as it prepares to go public.

The AI startup disclosed the customer moves in its S-1 filing for an initial public offering (IPO). The company has applied to list on the Nasdaq Global Select Market under the symbol “ANT,” but has not disclosed the number of shares or an expected price range.

The filing highlights a tension between Anthropic’s efforts to detect misuse and customers’ demand for data privacy.


“A substantial portion of our traffic is subject to Zero Data Retention (‘ZDR’) terms under which we retain only classifier scores, not prompts or outputs,” Anthropic said in the filing.

For that traffic, Anthropic said it cannot conduct offline or cross-conversation analysis, perform human reviews or examine historical data for possible misuse and security incidents.

“As our models become more capable, our safety framework has required and may continue to require monitoring and retention incompatible with ZDR for certain model classes,” the company said.

Anthropic said ZDR has become a competitive expectation for enterprise customers in regulated industries and for clients that access its models through third-party cloud platforms.

“Because ZDR is a competitive expectation for enterprise customers in regulated industries and on third-party cloud platforms, these requirements have caused and could continue to cause customers to migrate to other providers or to deployments where our visibility is more limited,” Anthropic said.

The filing does not name the customers, competing providers or affected models. It also does not disclose when the customers moved or how much revenue Anthropic lost.

Anthropic reported $16.3 billion in revenue and a net loss of $80.7 billion for the six months through June 2026, according to the filing. The company reported an operating loss of $693 million, compared with $2.1 billion a year earlier.

The company said in its IPO filing that more than 11,900 enterprise customers spend at least $100,000 a year on its Claude AI products. It added that as of September 30 more than 1,900 customers had spent $1 million or more over the past twelve months. About 250 had spent $10 million or more. The company also said it does not separately track “enterprise” customers. It uses the $100,000 threshold as a stand-in, because spending at that level generally reflects organisational rather than individual use.

In the filing, Anthropic said that it has delayed models and features, restricted availability and limited functions while conducting safety tests or adding safeguards. The company cited Mythos, which it initially released through the gated Project Glasswing programme, and said the decision “may have had a negative impact on our short-term revenue.”

The Claude-maker said its safety approach could also create “foregone or delayed revenue opportunities” compared with rivals that prioritise faster releases, broader access or more permissive uses. It added that such decisions may reduce adoption, limit customer usage or make its products less attractive.

"Customers, partners, or investors may not agree with or fully understand these decisions and may perceive them as limiting our competitiveness or commercial potential, which could adversely affect our reputation, customer relationships, market adoption, or valuation," Anthropic said.

"We cannot assure you that our long-term strategy will ultimately result in superior financial performance or that the anticipated long-term benefits of these decisions will be realised on the timeline we expect, or at all," it added.