OpenAI Unveils Private Safety Processing to One-Up Anthropic's Data Retention Policy
On August 19, 2026, OpenAI announced it is previewing a new privacy technology called Private Safety Processing for select customers. This automated system monitors for AI misuse with a key selling point: it watches for potential abuse while retaining none of the customer's data. The move is widely seen as a direct response to rival Anthropic, whose recently announced 30-day data retention policy has frustrated many enterprise customers.
Background: Anthropic's 30-Day Retention Policy Sparks Controversy
The story traces back to a policy Anthropic announced in July. For safety purposes, Anthropic said it would retain user data (including all sessions and the conversations within them) for a period of 30 days, applying to what it calls 'covered models' — including all Mythos-class models and 'future models with similar capabilities.'
The policy was designed to let the lab sift through and analyze potential impropriety. However, it has deeply concerned enterprises that handle large amounts of sensitive data and don't want it harbored — or inspected — by the AI lab. In an era of increasingly strict data compliance requirements, the policy has cost Anthropic goodwill among some enterprise customers.
OpenAI's Answer: How Private Safety Processing Works
OpenAI's new Private Safety Processing can be understood as a major upgrade to its existing Zero Data Retention (ZDR) policy. ZDR was already used in the OpenAI API: agents monitor for abuse on a per-session basis, customer data isn't retained by the company, yet malicious activity can still be scanned without human intervention.
OpenAI says Private Safety Processing is a new technology that widens ZDR's scope. It is described as a form of long-horizon safety monitoring that assesses the inputs and outputs of multiple conversations — not just one. The monitoring is conducted by an agent which, if triggered, catches interactions and analyzes them across sessions for signs of potential misuse.
The value of the new tech is detecting malicious use of AI that takes place over multiple sessions. As an OpenAI spokesperson told TechCrunch: a bad actor — hypothetically someone trying to engineer malware for a cyberattack — may spread out their requests to avoid detection. Private Safety Processing can analyze those multiple conversations for signs of abuse without human review of a user's conversations.
Trigger Mechanism and the Boundary of Human Review
In cases where the system is triggered, it may send a 'narrowly defined signal' to OpenAI that warns of a specific type of activity. Based on that signal, OpenAI can decide whether 'enforcement is necessary.' If so, OpenAI will reach out to the customer for more context or to work with them on the issue, and a customer may choose to share data with OpenAI at their discretion.
By contrast, Anthropic notes that human review of customer data can occur, but only 'through a controlled access path' involving 'a small set of approved reviewers.' Every review session is 'recorded in a tamper-proof log that reviewers cannot suppress or modify.' Both companies are trying to prove to enterprise customers that they value both safety and privacy — just through very different implementation paths.
Intensifying Rivalry: Privacy Becomes a Core Battleground
The corporate competition between OpenAI and Anthropic is tense at the moment, with both companies looking for any opportunity to gain an advantage. A recent report showed OpenAI's Q2 growth was slower than Anthropic's. Anthropic's annualized revenue run rate is now reportedly $65 billion, its investors have said it could IPO at a $2 trillion valuation, while OpenAI is also working on its own IPO.
For enterprise customers, this 'privacy arms race' is a good thing: it means leading AI vendors must make stronger data protection commitments to win enterprise deals. In the enterprise AI market, the balance between data sovereignty, compliance, and safety monitoring is becoming the key variable determining who wins the trust of large customers.
Sources
Frequently Asked Questions
Q1: What is Private Safety Processing?
A: It is a new technology OpenAI is previewing to select customers, an upgrade to its Zero Data Retention (ZDR) policy. It is an automated safety monitoring system that analyzes AI usage across multiple sessions to detect misuse, while retaining none of the customer's data and requiring no human review of conversations.
Q2: What is Anthropic's 30-day data retention policy?
A: Anthropic announced in July that, for safety purposes, it would retain user data (including all sessions and conversations) for 30 days, applying to 'covered models' including all Mythos-class models. The policy has frustrated enterprise customers handling sensitive data.
Q3: What does this mean for regular users?
A: The service is currently being previewed only for select enterprise customers. For regular users, the signal is clear: AI vendors are pushing 'zero data retention' capabilities into more complex cross-session scenarios, so enterprise customers can enjoy stronger data privacy guarantees while using AI.
Q4: Who will win the enterprise market, OpenAI or Anthropic?
A: It is too early to tell. Anthropic's annualized revenue run rate has reached $65 billion with faster growth, and investors expect a $2 trillion IPO valuation; OpenAI has reached $40 billion in revenue and is differentiating through privacy commitments. The balance between privacy and safety will be a key deciding factor.
🔧 Related AI Tool Recommendations
Detect whether text is AI-generated, supports multiple models
Quickly extract core content from long texts
Intelligently rewrite text, changing expression while preserving meaning
Conclusion
OpenAI's launch of Private Safety Processing is a landmark event in the escalating competition over enterprise data privacy in the AI industry. It shows that, as AI models become more powerful, 'how to ensure safety without invading privacy' has become a core battleground for AI vendors winning enterprise customers. OpenAI chose the technical path of 'zero data retention plus cross-session monitoring' to directly counter Anthropic's different approach to the safety-privacy tradeoff.
For enterprises, this competition brings more choices: companies that want AI capabilities but are extremely sensitive about data security now have options better suited to their needs. As both companies approach their IPOs, the interplay between privacy protection, safety monitoring, and customer trust will shape the enterprise AI market landscape for a long time.