New Anthropic safety system will let customers keep control of data

Anthropic is preparing to roll out a new safety system for its top models that could see data retention responsibilities handed back to enterprise customers, but a 30-day retention policy is likely to remain.

Technical Staff Member Sholto Douglas appeared to confirm reports on Thursday, posting on X that Anthropic would let safeguarding occur on customer infrastructure, the day after OpenAI detailed similar technology.

He said Anthropic had been working with more than 100 customers on a system where “Your data sits in infrastructure you own and control, and safeguards/monitoring is done via automated systems we provide to you.”

Anthropic and OpenAI believe the “responsible” way to provide top AI models involves monitoring output across requests “because anomalous activity is much easier to detect over hours or days of behaviour,” he added. “It looks like we've arrived at both the same conclusion - and the same solution.”

While Anthropic has not made an official announcement about the new system, Claude Code head Boris Cherney also posted on X to say it would arrive in the autumn.

Zero data retention?