OpenAI Previews Private Safety Processing to Counter Anthropic
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI is previewing "Private Safety Processing" to select customers, an automated agent-based system that monitors for potential abuse while retaining none of the customer's data.
- Anthropic's July data retention policy stores user sessions for 30 days for "covered models" — including all Mythos-class models, Fable, and future models with similar capabilities — a stance that has aggravated enterprise customers handling sensitive data.
- OpenAI says Private Safety Processing extends its existing Zero Data Retention approach by conducting long-horizon safety monitoring across multiple conversations, catching bad actors who spread malicious requests across sessions to evade single-session detection.
- Anthropic permits human review of retained customer data only "through a controlled access path" with a small set of approved reviewers, with every review session recorded in a tamper-proof log that reviewers cannot suppress or modify.
- When OpenAI's system flags activity, it sends a "narrowly defined signal" and may contact the customer for context; customers can choose at their discretion whether to share data with OpenAI.
- The competitive stakes are rising: Anthropic's annualized revenue run rate reportedly hit $65 billion — outpacing OpenAI's Q2 growth — with Anthropic investors suggesting a potential $2 trillion IPO as OpenAI also prepares to go public.
Why it matters: OpenAI is weaponizing Anthropic's safety policy against it, giving enterprises spooked by 30-day data retention a vendor that promises multi-session abuse detection while storing nothing. For customers handling sensitive or regulated data, the choice now pits Anthropic's stronger human-review safeguards against OpenAI's promise of total data non-retention.
Ask SkimNews
