OpenAI is testing Private Safety Processing, a new safety system for enterprise and API customers designed to detect multi-session misuse patterns without retaining customer data. The move directly counters rival Anthropic's unpopular policy of retaining data for 30 days on its Fable 5 and Mythos 5 models. Tech leaders like Satya Nadella and David Sacks have criticized data retention, making privacy a key battleground as OpenAI seeks to boost revenue growth relative to Anthropic.
Private Safety Processing system
- ▪OpenAI plans a broader rollout of Private Safety Processing and the publication of a technical white paper in September 2026
- ▪OpenAI announced on August 19, 2026, that it is testing a new automated safety system called Private Safety Processing with early enterprise and API customers
Zero data retention policy
- ▪OpenAI's Zero Data Retention controls do not apply to Free, Plus, Go, and Pro consumer ChatGPT subscription plans, whose data settings remain unchanged
- ▪Under the Private Safety Processing system, customer data can remain on customer-controlled infrastructure or be stored by OpenAI using encryption keys controlled by the customer
Long-horizon misuse detection
- ▪Private Safety Processing expands Zero Data Retention by using automated agents to analyze inputs and outputs across multiple conversations rather than monitoring interactions individually
- ▪When Private Safety Processing triggers a signal, OpenAI can request additional context from the customer, though the decision to share data remains at the customer's discretion
- ▪If Private Safety Processing detects suspicious behavior, it sends a narrowly defined safety signal to OpenAI without exposing the underlying prompts or responses
Enterprise customer privacy concerns
- ▪Anthropic's July 2026 policy requiring 30-day data retention for its Fable and Mythos models has concerned enterprise customers who handle large volumes of sensitive data
- ▪Tech leaders including David Sacks, Satya Nadella, and Alex Karp have criticized data retention policies, arguing that businesses in regulated sectors like healthcare and finance require data privacy certainty
OpenAI-Anthropic competitive positioning
- ▪OpenAI's launch of Private Safety Processing directly counters Anthropic's policy of retaining customer data for 30 days on its Fable 5 and Mythos 5 models to detect multi-request attacks
- ▪The privacy dispute comes amid intense competition where OpenAI's second-quarter revenue growth lagged behind Anthropic, whose annualized revenue run rate reportedly reached $65 billion
Story comments
Loading comments…