By Rajwa Quasim
In a race to stay on top of the increasingly competitive AI market, OpenAI is introducing a new privacy-focused safety system called Private Safety Processing.
Private Safety Processing is designed to allow OpenAI to identify potential misuse of its AI models without giving the company access to customers’ underlying data or retaining it. The company is previewing the service with select customers. The announcement comes amid growing pressure on AI companies over the misuse of their models and how they handle sensitive information.
Under the new approach, the system would analyze interactions for patterns that could indicate potentially harmful activity. The analysis would be conducted by an agent. For example, it could detect a bad actor trying to develop malware for a cyberattack who spreads requests across multiple conversations to avoid detection. Private Safety Processing can analyze those conversations for signs of abuse without requiring human review of a user’s conversations.
READ: OpenAI rolls out ChatGPT safety features for teens (August 18, 2026)
According to OpenAI, it will receive signals indicating such activity and take further action based on those signals. A company spokesperson said OpenAI may also contact the customer for more context or work with them to address the issue. Customers may choose to share their data with OpenAI at their discretion.
The Private Safety Processing system is broader than OpenAI’s Zero Data Retention policy. ZDR uses agents within the OpenAI API to monitor for abuse on a per-session basis. While customer data is not retained, the company can still scan for potentially harmful activity without human intervention.
This contrasts with Anthropic’s recent policy on safety measures. Its data-retention policy allows the company to retain user data, including sessions and conversations, for 30 days for covered models. These include all Mythos-class models and future models with similar capabilities. The policy, announced in June, has left some enterprises that handle large amounts of sensitive data unsettled or wary about having that information retained or reviewed by the company.
Anthropic uses Zero Data Retention except for Mythos models. The company said human review of data can occur only through a controlled system and by a small group of approved reviewers. It also said every review is recorded in a tamper-proof log that reviewers cannot suppress or modify.
READ: OpenAI files a motion to dismiss Apple lawsuit (August 6, 2026)
OpenAI is reportedly preparing for a potential initial public offering. The Wall Street Journal reported that OpenAI’s revenue grew to $6.7 billion in the three months ended in June, up from $5.7 billion in the first quarter. Meanwhile, its operating margin sank further into negative territory, pushing the company farther from profitability ahead of a much-anticipated IPO.
The company replaced its chief revenue officer, Denise Dresser, last week. To accelerate growth, OpenAI has also introduced a “super app” strategy focused on transforming ChatGPT from a basic chat interface into an all-in-one workspace. The strategy combines conversational AI with advanced coding capabilities from Codex and web-browsing automation to execute complex, multistep workflows.
Meanwhile, Anthropic’s revenue reached $11.6 billion in the second quarter. Its annual revenue is projected at $65 billion, and the company could pursue an IPO at a $2 trillion valuation.