Anthropic Builds a Privacy Shield for Business AI Conversations. You Are Probably Not Covered.

Key points

• Anthropic announced Enterprise Frontier Safeguards (EFS), a system that stores conversation data in customer-controlled cloud infrastructure instead of on Anthropic's own servers, rolling out in phases starting fall 2026.
• EFS was built with input from more than 100 enterprise customers across healthcare, finance, law and manufacturing, and works with AWS and Google Cloud.
• Until EFS is ready, eligible enterprise customers get zero data retention on Claude Fable 5 and Fable 5.1, meaning no conversation logs are kept.
• EFS covers Claude Code, Claude Enterprise, Amazon Bedrock and Google's Agent Platform, but ordinary consumer accounts are not included.
• Most consumer chatbot users still have close to no real privacy protection unless their employer holds a paid contract with an AI provider.

Every time you ask an AI chatbot a question, that conversation typically lands on a server owned by the company that built it. It can be used to improve the AI, handed over if a court demands it, or shared with partners. Anthropic's new Enterprise Frontier Safeguards, or EFS, tries to fix that, at least for business customers.

The idea is straightforward. Instead of your company's conversations travelling to Anthropic's infrastructure, they stay inside cloud storage your company already controls, whether that's AWS or Microsoft Azure. Anthropic pairs that arrangement with what it calls advanced safeguards for detecting misuse.

The catch is who gets it. EFS is an enterprise product. Your personal Claude account is not covered.

Why didn't Anthropic just offer zero data retention before?

It did, partly. But Anthropic's most capable models, the ones it calls "Mythos-class" such as Claude Fable 5.1, were previously excluded from zero data retention (ZDR, a contractual promise to delete any record of your conversation as soon as it finishes processing) because of misuse concerns. A smarter AI can do more damage in the wrong hands, including what Anthropic calls "autonomous misbehavior," where an AI agent, software that can carry out multi-step tasks on its own, acts harmfully without anyone explicitly ordering it to.

EFS is Anthropic's attempt to square that circle. The data stays in the customer's cloud, so Anthropic technically cannot see it. Abuse-detection still runs, but the customer organisation controls where the output lives.

Google and OpenAI have made similar moves. Google's AI memory product, announced last week, also promises it cannot read what it stores. OpenAI launched something called Private Safety Processing last month, which analyses for abuse on the customer's systems before deletion.

What does this mean for ordinary people?

Practically nothing, right now. EFS launches in phases starting later this fall, aimed squarely at companies, not individuals.

For most people using ChatGPT or Gemini at home, the privacy picture hasn't changed. Without a paid enterprise contract, the most you can count on is a policy promise, not a technical guarantee. Proxy services like DuckDuckGo's Duck.ai limit what the underlying AI sees about you, but your messages still reach that underlying model. Our coverage of what Anthropic may do with financial data inside Claude, published on 20 September, shows how far that gap between enterprise and consumer protection can stretch.

That gap matters. Cryptographer Moxie Marlinspike, who created the encrypted messaging app Signal, has said publicly that people now share more sensitive information with AI chatbots than they ever did in text messages, calling it several orders of magnitude more significant than the privacy risks messaging once posed.

The honest read on EFS: it's a real improvement for enterprises that previously had to choose between Anthropic's most powerful models and protecting their data. Anthropic built it alongside more than 100 customers across healthcare, financial services and law, which suggests the feature reflects actual business need. Whether the same level of protection ever reaches consumer accounts is the question Anthropic hasn't answered.

Common questions

Does EFS protect my personal Claude account?

No. EFS is for Claude Enterprise, Claude Code and developer platform accounts tied to cloud services like AWS and Google Cloud. Free and paid individual consumer accounts are not included in the initial rollout.

What is zero data retention, and do I have it?

Zero data retention, often shortened to ZDR, means the AI provider deletes any record of your conversation as soon as it finishes processing your request. Most individual users don't have this. It's typically only available through a paid enterprise contract between your employer and the AI provider.

Is my data safe if I use a privacy-focused proxy like Duck.ai?

Using a proxy reduces what the underlying AI model can tie back to your identity. But your messages still reach that underlying model, which has its own data practices. A proxy adds distance but isn't the same as a technical guarantee that no one can read your conversation.