AnswersPublished16 sources
Anthropic’s Enterprise Data Policy Is Changing Who Holds AI Safety Logs
Anthropic plans to keep a mandatory 30 day retention period for enterprise traffic from its most advanced models, but let customers store those logs in their own cloud infrastructure later in 2026. The policy was introduced in June to help detect misuse and cyberattacks whose patterns may emerge across multiple inte...
Published byEdited with GPT-5.6 LunaImages generated with GPT Image 1.5
Research answer
Key takeaways
- Anthropic plans to keep a mandatory 30 day retention period for enterprise traffic from its most advanced models, but let customers store those logs in their own cloud infrastructure later in 2026.
- The policy was introduced in June to help detect misuse and cyberattacks whose patterns may emerge across multiple interactions.
- OpenAI is pursuing a different, still previewed approach: Private Safety Processing is designed to identify misuse patterns without retaining the underlying customer content.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What changes is Anthropic planning to make to its enterprise data-retention policy and safety systems, including the requirement that busine. Article summary: Anthropic is planning to keep its 30-day retention requirement for business use of its most capable models, but change who holds the data: customers would be able to retain the required logs in their own cloud environmen. Topic tags: general, news, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts w
Anthropic is planning to change the location of enterprise AI safety logs—not eliminate the logs themselves. Business customers using the company’s most capable models would still face a 30-day retention requirement, but they would be able to keep the required data in cloud infrastructure they control rather than on Anthropic’s systems. The new system is expected later in 2026, although no specific launch date has been announced. 1
2
What Anthropic is changing
The proposed policy would give enterprises greater control over stored prompts and model outputs. Instead of Anthropic holding the required traffic on its own infrastructure, customers could retain it in their own cloud environments.
That distinction matters for organizations handling proprietary, regulated, or otherwise sensitive information. However, the planned change is not a zero-data-retention option: the relevant data would still need to remain available for 30 days for safety monitoring. 1
2
What stays the same
Anthropic’s 30-day rule would continue to cover traffic from its advanced Fable and Mythos models, as well as future frontier-class offerings. The company introduced the requirement in June, arguing that some emerging misuse—including potential cyberattacks—may only become visible when multiple interactions are examined together. 2
10
15
In practical terms, the proposed framework preserves three elements of Anthropic’s safety strategy:
- A defined retention window: covered prompts and outputs remain available for 30 days.
- Cross-interaction monitoring: safety systems can look for patterns that are difficult to identify from a single request.
- Enterprise-controlled storage: customers, rather than Anthropic alone, would manage where the retained data sits.
The final technical and contractual details remain limited because the system has not yet been fully rolled out or reflected in all product documentation. 2
7
Anthropic is reportedly testing the approach with customers
Anthropic has been developing the new system with more than 100 customers, including Salesforce, according to reporting cited in the available accounts. The reported collaboration suggests the company is trying to address enterprise concerns about data control while preserving the visibility it says is needed for safety work. 12
That trade-off is especially important for businesses that previously expected a zero-data-retention commitment to apply broadly across their Anthropic deployments. Under the proposed model, those customers could retain custody of the required logs, but the 30-day availability requirement for covered models would remain. 9
10
How OpenAI’s approach differs
OpenAI has previewed a different system called Private Safety Processing. It is designed to detect potential misuse patterns across related interactions while preserving zero-data-retention commitments, by sending a narrowly defined safety signal rather than retaining the underlying customer content. 9
10
The two approaches therefore make different architectural promises:
- Anthropic: retain covered traffic for 30 days, with the option for the enterprise to store it in its own cloud.
- OpenAI: attempt to identify safety risks without retaining the customer content itself.
OpenAI’s system is still a preview being tested with early customers, so its eventual scope and real-world effectiveness have not yet been established. Anthropic’s proposal is also reported rather than fully documented, making a direct production comparison premature. 2
10
What enterprise buyers should watch next
The central question is no longer only whether an AI provider retains data. It is also who controls the retained data, what access the provider has, and how safety signals are generated.
Before the Anthropic system becomes available, enterprise buyers will need clear answers on the storage architecture, access controls, deletion process after the 30-day period, and whether the arrangement satisfies their existing regulatory and contractual requirements. Until those details are published, the safest conclusion is straightforward: Anthropic is offering more control over retention, not removing the retention obligation.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Anthropic’s Enterprise Data Policy Is Changing Who Holds AI Safety Logs"?
Anthropic plans to keep a mandatory 30 day retention period for enterprise traffic from its most advanced models, but let customers store those logs in their own cloud infrastructure later in 2026.
What are the key points to validate first?
Anthropic plans to keep a mandatory 30 day retention period for enterprise traffic from its most advanced models, but let customers store those logs in their own cloud infrastructure later in 2026. The policy was introduced in June to help detect misuse and cyberattacks whose patterns may emerge across multiple interactions.
What should I do next in practice?
OpenAI is pursuing a different, still previewed approach: Private Safety Processing is designed to identify misuse patterns without retaining the underlying customer content.
Sources
reuters.comAnthropic plans to change enterprise data retention policy, ...
bloomberg.comAnthropic Plans to Change Data Retention Policy for Advanced AI
axios.comOpenAI previews zero-retention safety system as Anthropic ...
the-decoder.comAnthropic changes data retention policy after enterprise pushback