Anthropic announced Enterprise Frontier Safeguards on September 1, 2026, a new security program for corporate clients of Claude. Companies will be able to store their usage data in their own cloud infrastructure at AWS, Google Cloud, or Microsoft Azure, while automated analysis continues to detect suspicious patterns. The phased rollout begins in fall 2026, with no additional costs for the feature itself.
Companies retain control over location and keys
According to an announcement from Anthropic, the core of the innovation is the physical separation of usage data and Anthropic’s own infrastructure. Corporate clients will now store their interaction data in their own cloud accounts at Amazon Web Services, Google Cloud, or Microsoft Azure and manage encryption using their own keys. An automated analysis will continue to search the data for patterns indicating cyberattacks, stolen credentials, or autonomous misbehavior by AI agents, without Anthropic employees needing to view individual content.
All three components – own location, own keys, and automated review – are available as separate add-ons and are free according to Anthropic; only the usual storage and data transfer fees from the respective cloud provider will apply. There will be no changes to model behavior, billing through the API, or existing usage limits. Until the feature is fully rolled out, the option of zero storage introduced in June will remain for the models Fable 5 and Fable 5.1, where Anthropic retains no data at all. The integration extends to Claude Code, Claude Enterprise, the Claude platform, and partner environments such as Amazon Bedrock and Microsoft Foundry.
More than 100 companies are testing the new system
More than 100 companies from finance, healthcare, manufacturing, telecommunications, law, retail, and the public sector participated in the development, according to Anthropic, including about a quarter of the Fortune 100 companies. Eight major U.S. banks – including Goldman Sachs, Morgan Stanley, Citi, Bank of America, and Wells Fargo – tested the system through the joint analysis center Analysis and Resilience Center for Systemic Risk. Other partners mentioned by Anthropic include Mastercard, Visa, Salesforce, Snowflake, Stripe, Comcast, and KPMG.
With this, Anthropic delivers the official implementation of a plan that had already sparked discussions the previous month: In August, reports about a possible relaxation of the 30-day storage requirement introduced in June for the models Fable and Mythos had circulated – at that time unconfirmed, without product names and without a timeline. Enterprise Frontier Safeguards now names the procedure, lists the involved partners, and sets a concrete start date for fall 2026. The company has not yet commented on a possible launch in Europe or Germany.
Anthropic tightens security measures in parallel
The announcement comes during a phase of tightened internal security measures at Anthropic. As SecurityWeek reports, the British AI Security Institute had previously documented that the test model Claude Mythos 5 performed actions against real people and organizations during a security test without protective mechanisms after it unexpectedly gained access to the open internet. Anthropic subsequently introduced a classifier against sandbox breaches in real-time, blocked network access from test infrastructure by default, and deployed around 150 product engineers to security work.
According to the company, Enterprise Frontier Safeguards is an independent project, but it complements the same line: to technically demonstrate control over data and behavior of AI systems, rather than relying solely on internal review processes at Anthropic itself. For corporate clients in regulated industries, part of the responsibility for handling sensitive usage data is thus shifted back to their own cloud environment.
It will be crucial whether the automated abuse detection is fast enough in practice when Anthropic itself no longer has direct insight into customer data. The next test will be the broader rollout announced for fall 2026, where it will need to be shown how many of the over 100 pilot customers actually transition to their own cloud storage.
