Anthropic Fixed Its Data Retention Problem on September 1. The Fix Requires Your Own Cloud Account, So It Skips You
Anthropic announced Enterprise Frontier Safeguards on September 1, and for about ten minutes I thought it might finally answer the complaint I've had since June: Claude's frontier models retain your API traffic for 30 days whether you like it or not. It doesn't. EFS lets customers store those retention logs in their own AWS, Azure, or Google Cloud account instead of Anthropic's, with their own encryption keys and their own audit trail. That's a real fix, for a very specific kind of customer. I have an API key and a solo product. I am not that customer, and if you're reading this on a similar setup, neither are you.
What Anthropic actually shipped
The mechanics are straightforward once you read past the press release framing. Enterprise Frontier Safeguards stores activity data in cloud infrastructure the customer controls, not Anthropic's systems, and Anthropic doesn't charge for it. Your cloud provider bills you for the storage, reads, writes, and egress, the same way it bills you for any other bucket. Automated monitoring still runs against that data looking for signs of serious misuse (attempts at offensive cyber capability development, stolen credentials, that kind of thing), but the flags go straight to the customer's own security team. No Anthropic employee reviews it by default. Anthropic built this with feedback from more than 100 enterprise customers across financial services, healthcare, telecom, and the public sector, plus the Analysis and Resilience Center for Systemic Risk, whose members include the CISOs of Goldman Sachs, Morgan Stanley, Citi, Bank of America, and Wells Fargo. It rolls out in phases starting this fall.
That's the announcement. It's a genuinely good piece of engineering for the customers it's built for. It is also, by construction, not something you can turn on from a dashboard.
Why the June policy was a problem in the first place
Back in June, Anthropic introduced mandatory 30-day data retention alongside Fable 5, its stated reason being that catching sophisticated misuse, cyberattacks that unfold across many sessions and accounts, requires holding data long enough to correlate patterns over time. Anthropic has been consistent that this data isn't used for training without explicit permission. The problem was never really about training. It was that regulated industries have their own rules about who is allowed to hold sensitive logs, and "a third party we don't operate" doesn't clear that bar no matter how good the stated intentions are. That's a legitimate objection if you're a bank. It's a much smaller concern if you're one person running a Claude-powered app for a few thousand users, where the retention policy is mostly an annoyance you flagged in a privacy policy and moved on from.
Who gets the fix, and who doesn't
Enterprise Frontier Safeguards piggybacks on Anthropic's existing Zero Data Retention program, and that program has never been self-serve. According to Anthropic's own platform docs, ZDR is "enabled on a per-organization basis" and each new organization needs it turned on separately by an Anthropic account team, after you contact sales to confirm eligibility. EFS adds another layer on top of that: you need actual cloud infrastructure of your own to receive the logs into, which means someone on your side has to stand up an S3 bucket or Blob container, wire up the encryption keys, and build the review workflow for the flags Anthropic's automated system sends you. That's a project, not a setting. A solo developer building a SaaS tool on the Claude API doesn't have a security team to review flags, doesn't have a dedicated cloud account earmarked for compliance logging, and almost certainly isn't going to get an Anthropic account rep on the phone to negotiate ZDR eligibility for a single-founder business. The 30-day retention rule from June stands for you exactly as it did the day it shipped.
The part worth sitting with
I don't think Anthropic did anything wrong here. Building EFS for banks and healthcare companies first makes sense: those are the customers who can't use the product at all without it, and they're the ones with the leverage and the compliance headcount to co-design something like this over months. The framing in the coverage, though, treats this as a resolution to "the retention policy problem," full stop, and that's where I'd push back. It resolves the problem for enterprise accounts. If you're a solo operator whose whole compliance story is a privacy policy page and a good-faith reading of Anthropic's usage policy, nothing changed for you on September 1. The retention clock started in June and it's still running exactly the same way.
The honest take
If data retention is genuinely a blocker for your product, an EU-only customer base that objects to a US company holding session logs, for instance, don't wait for a trickle-down version of EFS to reach individual developers. It's not obviously coming, and Anthropic has no stated plan to extend it below the enterprise tier. Your actual options are to route sensitive conversations away from the frontier models entirely, disclose the retention terms clearly to users and let them opt out of sensitive usage, or accept the 30-day window as a cost of using the best available model and build your compliance story around that fact rather than around a feature you can't access. Where I could be wrong: Anthropic explicitly said EFS integrates with Claude Code, the Claude Platform, and cloud-partner offerings across AWS, Google Cloud, and Microsoft Azure, and said it's "working to support third-party offerings that serve customers eligible for Enterprise Frontier Safeguards." If that eligibility bar turns out to be lower than "100+ regulated enterprises," and some cloud marketplace reseller starts offering a lightweight version to smaller accounts, this piece ages fast. I'd take that bet against myself happily.
Author
Lukas
@lukcombinator