Anthropic Reverses Data Retention Policy, Introduces Enterprise-Grade Security Alternative
nashnova research
Anthropic on Tuesday scrapped the mandatory 30-day data retention policy that triggered widespread enterprise pushback, replacing it with a programme called Enterprise Frontier Safeguards — a concession that, with a large-scale IPO widely expected, signals keeping big clients matters more than holding the line.
What made the original policy so unpopular?
In June, Anthropic launched its flagship models Claude Fable 5 and Mythos 5 alongside a mandatory 30-day retention rule: all traffic using those models had to be stored.
The stated rationale was preventing misuse and "sophisticated novel" cyberattacks; the company pledged not to use retained data for model training.
But the core enterprise concern was not *how* the data would be used — it was that the data left their perimeter at all. This means → no matter what Anthropic promised, data outside the customer's own walls is a risk in itself.
What does the new plan actually change?
The replacement is called Enterprise Frontier Safeguards. The key shift: enterprises regain control over how their data is reviewed, stored, and managed.
Customers can enable automated safety monitoring with no Anthropic human involvement required.
In plain terms = the old model was "hand us your data and we'll watch it for you." The new one is "here's the scanning tool — run it inside your own system; we never touch the data."
The programme carries no extra charge, works whether clients connect to Anthropic directly or via cloud providers, and is set for broader rollout this autumn.
Why did Anthropic back down?
Enterprise clients are Anthropic's primary revenue source, and the market widely expects a large-scale IPO.
This means → in IPO valuation logic, enterprise retention rate is one of the hardest numbers; a single data policy that drives clients away costs far more than the security benefit it provides.
Americas head Kate Jensen said the company spent hundreds of hours co-developing the alternative with clients — this reflects pushback at a scale well beyond "a few complaint emails."
Where does the rivalry with OpenAI stand now?
The policy reversal comes after OpenAI launched its own data-retention solution earlier this month; competition between the two for enterprise business continues to intensify.
In plain terms = OpenAI gave enterprises a "we won't touch your data" option first. If Anthropic hadn't matched it, clients had a concrete reason to switch.
Whether Anthropic can hold its large-client base before an IPO will be a critical test of its valuation thesis.
市场有风险，内容仅供研究参考，不构成投资建议。