Table of Contents
Anthropic data storage is set to become more flexible for enterprise customers, with the Claude maker reportedly developing a system that would allow businesses to retain their AI data on their own cloud infrastructure.
According to Bloomberg, Anthropic plans to introduce the option later this year as part of a new safety system. Enterprise customers would still be required to retain certain data for 30 days, but they would reportedly have the option to store it within their own cloud environments rather than exclusively on Anthropic’s infrastructure.
The reported change comes as businesses, particularly those in regulated industries, have raised concerns about sharing sensitive information with AI providers.
Why Anthropic Changed Its Data Policy
Anthropic announced in June that it would require 30-day retention of prompts and outputs from its most powerful Mythos and Fable models to help detect and prevent cyberattacks.
The policy represented a significant change for customers that had previously relied on zero-retention arrangements.
Anthropic acknowledged in a recent risk report that the policy could be unpopular with customers and create business risks, especially if competing AI providers maintained more privacy-focused policies.
According to Bloomberg, Anthropic has been coordinating with more than 100 customers in highly regulated industries while developing the new option.
Customer-Controlled Anthropic Data Storage
Under the reported plan, enterprise customers would continue meeting Anthropic’s 30-day retention requirement while gaining greater control over where the information is stored.
Rather than requiring all retained information to remain on Anthropic’s infrastructure, organizations could reportedly store the data within their own cloud computing environments.
This could be especially relevant for companies with strict regulatory, privacy and data-governance requirements.
However, the technical details of the planned system have not yet been publicly disclosed. Anthropic declined to comment on the reported plans.
OpenAI Is Taking a Different Approach
The discussion around Anthropic data storage comes shortly after OpenAI announced a different approach to AI safety monitoring.
OpenAI is testing a system called Private Safety Processing, which is designed to identify potentially harmful activity across multiple interactions without retaining customers’ underlying prompts and responses.
OpenAI says the system sends only narrowly defined safety signals to the company rather than exposing the raw conversation data.
The company plans a broader rollout and a technical white paper in September. Early customers reportedly include Databricks and Microsoft.
Two Different Approaches to AI Safety
The contrasting strategies highlight an important debate over how AI companies should balance security monitoring with data privacy.
Anthropic’s approach is based on the idea that retaining relevant information for a limited period can help identify sophisticated attacks that may only become apparent when multiple interactions are examined together. Giving customers control over storage could potentially address some concerns about sensitive information being held by the AI provider.
OpenAI, meanwhile, is exploring whether cryptographic techniques can support safety monitoring without giving the company access to customers’ underlying conversations.
Neither approach can currently be considered universally superior. Their effectiveness will depend on implementation, security controls, regulatory requirements and individual enterprise needs.
What This Means for Enterprise AI Buyers
For organizations adopting AI, data governance is becoming an increasingly important consideration.
Companies in healthcare, finance, government and other highly regulated sectors often need strict controls over where sensitive information is stored and who can access it.
A customer-controlled Anthropic data storage option could make Claude’s enterprise offerings more attractive to organizations that want greater control over their information while continuing to use advanced AI models.
OpenAI’s privacy-focused safety system could appeal to organizations that would rather avoid retaining raw prompts and responses altogether.
Key Details
| Category | Anthropic | OpenAI |
| Safety approach | 30-day data retention | Private Safety Processing |
| Storage approach | Planned customer-controlled cloud option | Designed to avoid retaining raw customer data |
| Purpose | Detect and prevent sophisticated cyberattacks | Identify misuse patterns across interactions |
| Enterprise involvement | More than 100 regulated-industry customers reportedly involved | Early customers include Databricks and Microsoft |
| Current status | Reported planned rollout later this year | Testing underway |
FAQs
What is Anthropic data storage?
The term refers to how Anthropic stores and manages information associated with its AI services. The company is reportedly developing an option that would allow enterprise customers to retain certain required data on their own cloud infrastructure.
Why is Anthropic introducing customer-controlled storage?
The reported plan comes amid concerns from enterprise customers about retaining sensitive prompts and outputs on an AI provider’s infrastructure.
Will Anthropic still require data retention?
According to the report, the planned system would still require enterprise customers to retain certain data for 30 days, but customers could potentially control where that information is stored.
How is OpenAI’s approach different?
OpenAI is testing Private Safety Processing, which aims to identify safety risks across interactions without retaining customers’ underlying prompts and responses.
Conclusion
The reported changes to Anthropic data storage reflect a broader shift in enterprise AI, where companies increasingly need to balance powerful AI capabilities with strict privacy and security requirements.
Anthropic’s planned customer-controlled storage option could give businesses greater control over retained information while supporting the company’s approach to detecting sophisticated cyber threats. OpenAI is pursuing a different path by exploring safety monitoring that does not require retaining raw customer conversations.
As enterprise AI adoption grows, data storage, privacy and security policies are likely to become increasingly important factors when organizations choose between competing AI platforms.