{
  "id": 2117466,
  "title": "OpenAI wants to monitor AI abuse without forcing customers to hand over their data",
  "url": "https://urgent.news/2026/08/20/openai-wants-to-monitor-ai-abuse-without-forcing-customers-to-hand",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-20T10:42:12.000Z",
  "source": {
    "name": "Digital Trends",
    "slug": "digital-trends",
    "url": "https://www.digitaltrends.com/computing/openai-wants-to-monitor-ai-abuse-without-forcing-customers-to-hand-over-their-data/"
  },
  "original_language": "en",
  "account": "OpenAI aims to enhance its capability to identify misuse of its advanced AI models without requiring customers who prioritize privacy to relinquish their data. In an announcement on August 19, OpenAI unveiled Private Safety Processing, a system intended to detect anomalous behavior across various interactions while maintaining the Zero Data Retention (ZDR) option for suitable API customers. OpenAI explains that its standard ZDR safeguards evaluate each request separately, which makes it challenging to detect patterns spread over multiple interactions. Private Safety Processing is intended to address this gap without granting OpenAI access to the original queries or responses. Currently, the system is being piloted with initial customers. Reports suggest that it targets enterprise and API clients rather than individual ChatGPT users. Rather than examining each request individually, Private Safety Processing can identify suspicious patterns across related interactions. When potential misuse is detected, OpenAI receives only a limited indication of the activity, not the customer's original content. With Zero Data Retention implementations, that content remains on infrastructure controlled by the customer. OpenAI is also creating another setup where encrypted content can reside on its infrastructure while customers retain the encryption keys. OpenAI itself would not possess those keys. Anthropic has adopted a different approach. For its Covered Models, prompts and outputs are stored for 30 days, including for organizations that previously utilized Zero Data Retention. Anthropic argues that this retention period aids its safeguards in catching attacks that only become apparent after several requests. OpenAI aims to tackle the same safety issue while preserving ZDR for eligible customers. For businesses that adopted ZDR specifically to limit how long an AI provider can retain their content, OpenAI's approach presents a materially different proposition. Private Safety Processing is still an early preview, and OpenAI has not yet publicly demonstrated that it can provide abuse detection equivalent to methods that retain customer content. The technical specifics of these privacy protections are not publicly available yet. There is an important exception. OpenAI states that images flagged as potential child sexual abuse material can still be retained for legally mandated manual review and reporting, even under Zero Data Retention. OpenAI plans to begin rolling out Private Safety Processing and publish a technical white paper in September. This should provide customers with a clearer understanding of whether cross-session safety monitoring can function without compromising the existing ZDR protections they rely on. Google currently lags behind OpenAI and Anthropic at the forefront of the frontier-model race. While Anthropic's Claude Opus 5 and OpenAI's GPT-5.6 Sol outperform Google's Gemini 3.1 Pro on major independent benchmarks, Google's Gemini 3.5 Pro remains in testing after reportedly failing to meet internal performance goals. As Google strives to narrow the gap, it is simultaneously marketing Gemini aggressively to a distinct audience. Eligible U.S. college students can now access Google AI Pro free for an entire year, along with a new Student Hub tailored for studying, coursework, and research. Google has significantly upgraded Gemini Live with the addition of Deep Research, a feature now available to all Gemini app users. Deep Research allows users to initiate a research task through voice command and return to the completed report later. Unlike previous chatbot responses, Deep Research can search multiple sources, follow relevant leads, and compile a more detailed report with citations. Although the process may take several minutes, users do not need to keep Gemini open while it works. Once the research is finished, users receive a notification. Meta has introduced its Meta AI assistant to the Mac through a dedicated desktop application, offering users another AI chatbot to keep open alongside existing tabs. The app can analyze shared windows, answer questions about screen content, generate content, and accept voice dictation across macOS.",
  "summary": "OpenAI’s Private Safety Processing promises to detect abuse across multiple interactions while preserving Zero Data Retention, putting its privacy strategy in direct contrast with Anthropic’s 30-day retention requirement.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Computerworld",
        "title": "OpenAI ‘temporarily slows’ scaling efforts, also promises zero data retention for select frontier model customers",
        "url": "https://urgent.news/2026/08/20/openai-temporarily-slows-scaling-efforts-also-promises-zero-data",
        "published": "2026-08-20T00:08:33.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}