How does X/XAI’s process for reporting CSAM to NCMEC work and what numbers have they publicly provided in 2024–2025?
Executive summary
OpenAI says it detects suspected child sexual abuse material (CSAM) in user content and reports confirmed instances — including uploads and user requests — to the National Center for Missing & Exploited Children (NCMEC) via the CyberTipline, and its transparency updates show a steep rise in the volume of those reports between 2024 and 2025 [1] [2].
1. How the reporting channel works: CyberTipline and ESP obligations
Electronic Service Providers (ESPs) are required under federal law to send suspected CSAM to NCMEC’s CyberTipline, and industry reporting consistently describes the CyberTipline as the mechanism through which platforms forward case details, hashes, and preservation requests to NCMEC and law enforcement [2] [3].
2. What OpenAI/X says it reports and how it classifies items
OpenAI has stated publicly that it reports all instances it deems to be CSAM — encompassing both content uploads and user requests that generate or attempt to generate CSAM — and that its safety teams are constantly improving detection tooling to surface and forward confirmed material to relevant authorities including NCMEC [1].
3. The numbers OpenAI has publicly provided for 2024–2025
OpenAI’s public disclosures — summarized in recent reporting — say the company sent 947 CyberTipline reports covering roughly 3,252 pieces of content in the first half of 2024, and that in the first half of 2025 it sent roughly 80 times as many child exploitation incident reports as it did in the comparable 2024 period, though OpenAI’s update and secondary reporting do not publish a full twelve‑month 2025 tally in the available reporting [1] [4].
4. Where those OpenAI figures sit relative to the broader CyberTipline totals
NCMEC’s public CyberTipline data shows dramatic sector-level trends: the center reported a 1,325% increase in reports involving generative AI between 2023 and 2024 (from 4,700 to 67,000), and NCMEC’s 2024 dataset and ESP breakdowns show millions of CyberTip reports overall — for example, roughly 20–20.5 million ESP-sourced reports in 2024 depending on the dataset and presentation — providing context that OpenAI’s several‑thousand‑item figures are a small slice of a very large ecosystem of platform reporting [5] [6] [7].
5. Why year‑to‑year counts jump and why they can be hard to compare
Part of the apparent volatility reflects both real increases in AI‑related imagery and changes in reporting mechanics: NCMEC introduced “bundling” to collapse duplicate tips tied to viral incidents, Congress enacted the REPORT Act in 2024 expanding what must be reported, and platforms have different detection tooling and thresholds — all of which make raw counts across years and companies difficult to compare without careful methodological notes [7] [3] [8].
6. How other labs and platforms report for perspective
Other AI and technology companies publish their own tallies — Microsoft reported hundreds of thousands of items actioned and reported to NCMEC in late‑2024 reporting periods and Anthropic disclosed it detected and reported 859 images to NCMEC between April 15, 2024 and March 31, 2025 — illustrating both variance in detection approaches and the fact that firms are publicly disclosing different metrics and timeframes [9] [10].
7. Caveats, competing narratives and potential agendas
Reporting aggregations and media accounts can amplify a single company’s percentage increase without always clarifying baselines or absolute scale, which risks overstating one firm’s share of the total problem or, conversely, understating systemic gaps; industry transparency efforts, advocacy groups and regulators all have incentives — either to highlight progress or to press for stronger rules — so numbers are often used to advance policy aims as much as to inform the public [1] [5] [3].
8. Bottom line: process is standardized, the numbers are fragmented
The mechanics are straightforward: platforms detect suspected CSAM, preserve evidence and submit CyberTip reports to NCMEC as required by law, but public numbers for 2024–2025 are fragmented across company updates, NCMEC aggregate reports and new bundling rules, with OpenAI publicly noting hundreds-to‑low‑thousands of reports in early 2024 and a proclaimed large multiple increase into 2025 while NCMEC-level data shows a larger wave of AI‑related reports industry‑wide [2] [1] [5].