OpenAI aims to track AI misuse without requiring customers to provide their data.

      OpenAI's new Private Safety Processing system adopts a different strategy than Anthropic, which mandates a 30-day data retention period for certain advanced models.

      OpenAI aims to improve its ability to identify misuse of its advanced AI models without requiring privacy-minded customers to relinquish Zero Data Retention (ZDR). In an announcement on August 19, OpenAI unveiled Private Safety Processing, a system intended to detect suspicious activities across various interactions while maintaining ZDR for qualifying API clients.

      OpenAI explains that its current ZDR measures analyze requests on a case-by-case basis, making it difficult to identify patterns from multiple interactions. Private Safety Processing is designed to bridge this gap without allowing OpenAI staff to access the original prompts or responses.

      Currently, the system is undergoing testing with select early adopters. According to Axios, it primarily targets enterprise and API customers instead of individual ChatGPT users.

      Rachit Agarwal / Digital Trends

      How Private Safety Processing functions

      Rather than examining requests individually, this system can identify suspicious patterns across related engagements. If it detects potential misuse, OpenAI gets a limited signal indicating the occurrence, rather than accessing the customer's original content.

      In scenarios involving Zero Data Retention, that content remains on infrastructure managed by the customer. OpenAI is also working on an alternative configuration where encrypted content can be stored on its infrastructure while customers retain the encryption keys, which OpenAI would not have access to.

      Why Anthropic took a different path

      Anthropic opted for a different compromise. For its Covered Models, both prompts and outputs are stored for 30 days, even for organizations that previously utilized Zero Data Retention. Anthropic asserts that this timeframe allows its safeguards to identify attacks that may only be evident through multiple requests.

      OpenAI seeks to address the same safety challenge while preserving ZDR for eligible clients. For businesses that implemented ZDR specifically to control how long an AI provider retains their content, the two approaches are significantly different.

      Aerps.com / Unsplash

      What zero retention still excludes

      Private Safety Processing remains an early preview, and OpenAI has not publicly demonstrated its capability to provide abuse detection comparable to methods that retain customer content. The technical specifics of these privacy measures are not yet available to the public.

      There is one clear exception: OpenAI states that images identified as potential child sexual abuse material can still be retained for required manual review and reporting, even within the Zero Data Retention framework.

      OpenAI intends to begin the rollout of Private Safety Processing and release a technical white paper in September, which should provide customers with a clearer understanding of whether cross-session safety monitoring can be effective without compromising the current ZDR protections they depend on.

      Paulo Vargas is an English major turned reporter turned technical writer, with a career that has consistently circled back to…

      The CIA connection that aided NeXT's survival and Jobs’ return to Apple

      The CIA may have played a role in keeping Steve Jobs’ NeXT afloat.

      The narrative of Steve Jobs’ return to Apple is often considered one of Silicon Valley's success stories. A recent account suggests there was another, much less public chapter that contributed to this outcome: a secret relationship between Jobs’ struggling computer company, NeXT, and the CIA. According to a report by Sharon Weinberger in The Wall Street Journal, the CIA quietly became a significant customer for NeXT in the late 1980s, purchasing thousands of specially configured workstations when Jobs' company was in dire need of sales. The report claims this arrangement helped ensure NeXT's survival and provided Jobs with additional years to grow as a business leader before Apple rehired him in 1996.

      Read more

      Windows 11 is eliminating the Drag Tray feature

      Windows 11’s troublesome Drag Tray is finally being discontinued.

      Microsoft is discontinuing one of Windows 11's more peculiar file-sharing experiments after users indicated that the feature was often more frustrating than beneficial. The Drag Tray was meant to offer a quicker method for sharing files. By dragging a file from File Explorer to the upper part of the screen, a floating panel would appear with shortcuts to send it to apps. While theoretically simple, many users were already dragging files to that area of the screen for other reasons.

      Read more

      Google is providing students a complimentary year of AI Pro and a new Gemini Student Hub

      Gemini is getting additional features to assist students in studying, organizing, and researching.

      Google currently trails OpenAI and Anthropic in the upper echelons of the frontier-model race. Anthropic’s Claude Opus 5 and OpenAI’s GPT-5.6 Sol both surpass Gemini 3.1 Pro on major independent evaluations, while Google’s anticipated Gemini 3.5 Pro remains in testing after reportedly not meeting internal performance benchmarks. As Google strives to close this gap, it is also aggressively directing Gemini toward a different demographic. Eligible U.S. college students can now access Google AI Pro at no cost for an entire year, in addition to a new Student Hub specifically designed for studying