- Published on
OpenAI seeks to one-up Anthropic with new customer privacy protections
- Authors
- Name
- DeepHorus
As artificial intelligence models grow in capability, so does the risk of misuse. AI companies now face a delicate tightrope between respecting enterprise customer privacy and monitoring for potential abuse. Sensing an opportunity to one-up its rival Anthropic, OpenAI has announced a privacy-centric safety approach called Private Safety Processing. The company is previewing the new service to select customers, positioning it as a way to watch for potential abuse while retaining none of the customer’s data.
This system directly counters Anthropic’s recently announced data-retention policy, which has aggravated some customers. That policy, introduced in July, requires the AI lab to keep user data—including all sessions and conversations—for 30 days on "covered models" such as Mythos-class models and future models with similar capabilities. Anthropic designed this policy for safety purposes, allowing the lab to sift and analyze potential impropriety. However, it deeply concerns enterprises that handle large amounts of sensitive data and do not want it harbored or inspected by an AI lab.
OpenAI, like most other AI companies, already provides customers with a relative level of privacy through Zero Data Retention (ZDR). ZDR uses agents within the OpenAI API to monitor for abuse on a per-session basis, so customer data is not retained while bad activity is still scanned without human intervention. Anthropic also largely abides by ZDR, except for its "covered models." Now, OpenAI says Private Safety Processing is a new technology that widens ZDR’s scope.
According to OpenAI, Private Safety Processing is 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. This helps detect malicious use of AI that takes place over multiple sessions. For example, a bad actor hypothetically trying to engineer malware for a cyberattack might spread out requests to avoid detection; the new system can analyze those conversations for abuse without human review.
In the case where the system is triggered, it may send a "narrowly defined signal" to OpenAI, warning 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 together on the issue, and the customer may choose to share data at their discretion. This approach avoids any long-term storage of customer data, addressing a major pain point for privacy-sensitive enterprises.
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. While this transparency is commendable, the fact remains that Anthropic stores and potentially inspects customer data for covered models. OpenAI’s new system eliminates that requirement entirely, making it an attractive option for businesses in regulated industries like finance, healthcare, and law.
The timing is significant, as the corporate competition between OpenAI and Anthropic is tense. A recent report showed OpenAI’s Q2 growth was slower than Anthropic’s, and Anthropic’s annualized revenue run rate is now reportedly $65 billion. Anthropic investors have suggested it could IPO at $2 trillion, while OpenAI is also working on its own IPO. In this environment, privacy protections could become a key differentiator in winning enterprise customers, a critical revenue segment for both companies.
Looking ahead, OpenAI’s move may pressure Anthropic to rethink its data retention policy or develop similar privacy-preserving monitoring mechanisms. It also underscores a broader industry trend: as AI becomes more powerful, companies must innovate to balance safety, privacy, and regulatory compliance. Private Safety Processing is still in preview, but if it proves effective, it could set a new standard for how AI vendors handle customer data. Enterprises will be watching closely, as will regulators and competitors. The future of AI safety is not just about stopping misuse—it’s about doing so without compromising trust.
