Anthropic Lets Enterprise Customers Store Their Own AI Data on Own Cloud Infrastructure

Anthropic Revises Data Retention Policy for Enterprise Customers

Anthropic is changing its data retention policy to allow enterprise customers greater control over how they store their own AI data1

1

2

2

. The Claude chatbot maker will still require business customers to retain data for 30 days, but now gives them the option to keep it on their own cloud infrastructure rather than on Anthropic's systems4

4

. This shift addresses mounting concerns from enterprise customers about data control options, particularly those operating in highly regulated industries3

3

.

The company plans to roll out this new safety system later this year. The proposed changes have been in development for months, with Anthropic coordinating with over 100 customers, including Salesforce, to develop the system2

2

. The updated enterprise data retention policy applies to Anthropic's advanced AI models including Fable and Mythos, as well as future frontier models1

1

.

Why the 30-Day Data Retention Requirement Sparked Concern

Anthropic announced the 30-day data retention requirement in June, citing the need to guard against potential cyber threats using its technology1

1

. The company stated that the data would help defend against complex and novel attacks, including new jailbreaks and attacks that operate across many requests, as well as help identify and reduce false positives3

3

.

However, the policy raised immediate red flags. Microsoft reportedly limited employees' use of Claude Fable 5 while its legal teams evaluated the changes to Anthropic's data retention requirements3

3

. The retention policy also included keeping prompts and outputs for up to two years if they were flagged by Anthropic's trust and safety classifiers as violating the company's usage policy3

3

. This created questions in C-suite and CFO offices about what exactly happens to data flowing through enterprise AI systems, how long it is retained, and what liabilities that creates.

How Anthropic's Approach Compares to OpenAI's AI Safety Strategy

The timing of Anthropic's policy shift is notable. Just one day before this announcement, rival OpenAI unveiled a safety system capable of flagging potential misuse risks of its technology without holding on to customer data1

1

. OpenAI announced it is testing Private Safety Processing, which identifies serious safety risks that may only become visible across multiple interactions, while continuing to offer Zero Data Retention for eligible API customers3

3

. OpenAI plans to start rolling out Private Safety Processing in September.

This competitive dynamic highlights the growing tension between AI safety requirements and enterprise data sovereignty concerns. Both companies are racing to demonstrate they can balance AI governance with customer data control, but they're taking different technical approaches to achieve that balance.

Enterprise AI Spending Under Pressure

The policy change comes as enterprise AI spending faces increased scrutiny. Some businesses have been cutting back on Anthropic and OpenAI products in favor of cheaper alternatives, demanding clearer returns on investment1

1

. Flo Crivello, CEO of AI startup Lindy, moved all of his company's traffic from Anthropic's Claude models to DeepSeek, a Chinese company that makes lower-cost, open-weight alternatives, saying the switch would save Lindy millions of dollars within months1

1

.

Despite these pressures, Anthropic's financial performance remains strong. The company's annualized revenue run rate reached $47 billion in May, according to CNBC1

1

. Both Anthropic and OpenAI filed confidential IPO prospectuses with the Securities and Exchange Commission in early June1

1

.

Watch for how this policy evolution affects enterprise adoption rates in sectors like healthcare, finance, and government, where data residency requirements are non-negotiable. The ability to store their own AI data on own cloud infrastructure may prove decisive for organizations that have been hesitant to adopt advanced AI models due to compliance concerns. As AI safety and data sovereignty continue to collide, expect more providers to offer hybrid approaches that balance security monitoring with customer control.