Anthropic said Tuesday it will replace a controversial data retention policy for business customers with a new system called Enterprise Frontier Safeguards after "a lot of feedback." The company said the new setup will let businesses control how their data is reviewed, stored and managed, while still allowing automated safety monitoring without any Anthropic human review. That’s the kind of corporate fix that arrives only after customers push back hard enough to make the old arrangement harder to defend.
Who Gets to Control the Data
The company said it will not charge for the safeguards, and that they will work whether customers use its technology directly or through a cloud provider. Kate Jensen, head of Americas at Anthropic, said the company spent hundreds of hours working with customers to develop the new approach. "We can do the scanning that we really need to, to help everybody feel confident, but it's on their systems and their data so that they don't have to compromise on their own privacy posture," Jensen said. She said she feels "very confident" Anthropic will see accelerated deployment after the announcement, especially on customers' most sensitive workloads.
That language does a lot of work. The company keeps the scanning. The customers keep the data. Anthropic keeps the power to monitor, but only after the enterprise market made clear it wasn’t eager to hand over its traffic without a fight. The arrangement is framed as privacy protection, but it also shows how much leverage the biggest clients can exert when a platform’s business model depends on them.
The June Policy and the Backlash
Anthropic introduced the data retention policy in June alongside Claude Fable 5 and Mythos 5, two of its most advanced artificial intelligence models. At the time, the company said it would require 30-day retention for all traffic on those models to help defend against misuse and "complex and novel" cyberattacks. Anthropic said it would not use that data for any "non-safety-related purpose," including model training, but many enterprise users raised concerns. The June policy still applies to non-enterprise subscribers using its Mythos-class models.
That split matters. Enterprise customers got a revised system after objecting, while non-enterprise subscribers remain under the original retention rule. Same company. Different treatment. The hierarchy is plain enough: those with more bargaining power get the softer terms, and everyone else stays inside the old apparatus.
Founded in 2021, Anthropic said the enterprise market generates the vast majority of its revenue and is crucial as it prepares for what is widely expected to be a massive IPO. The company is also racing to stay ahead of rivals like OpenAI, which introduced its own data retention solution earlier this month. The pressure here isn’t some abstract public good. It’s revenue, competition and the march toward a public offering, with customer data sitting in the middle of the scramble.
What the Top Wants
Last week, Anthropic CEO Dario Amodei previewed Enterprise Frontier Safeguards during an interview with CNBC about the company's expanded partnership with Salesforce. Amodei said the solution helps ensure that Salesforce's customer data remains private and that Anthropic's models "don't run out of control." He said Anthropic put a "huge amount of effort" into managing permissions for the integration between the two companies.
The phrasing is revealing. The company wants to reassure enterprise clients that the machine won’t wander off the leash, while also making clear that permissions, monitoring and control remain central to the deal. The safeguards may be free, but they’re still part of a system built to keep corporate customers inside the platform and the platform inside their operations.
Anthropic also unveiled Claude Fable 5.1 and Mythos 5.1 on Tuesday, saying the new models show improved coding, knowledge work and research capabilities. Enterprise Frontier Safeguards will roll out in phases, and Anthropic said it is aiming for broader availability this fall.
Separately, the U.S. government signaled a preference for a hands-off approach to AI regulation at the G-20, according to officials. That posture leaves the field to the companies themselves, the same companies already racing for market share, IPO value and control over the systems people are told to trust. When the state steps back, the bosses don’t disappear. They just move faster.