Raw Story & AlterNet v. OpenAI

The allegations

Raw Story, AlterNet and, in a separate same-day filing, The Intercept sued OpenAI on a narrower theory than most AI publisher suits: not that training itself infringed their copyrights, but that OpenAI stripped bylines, titles and copyright notices — copyright-management information, in the statute's terms — from their articles before feeding them into ChatGPT's training data. The outlets hadn't registered the articles at issue, which forecloses an ordinary infringement claim, so the DMCA's CMI-removal provision was the available vehicle. It's a narrower claim, but also a cheaper one to bring: it doesn't require proving the underlying work was infringed, only that management information was intentionally stripped.

The legal question

The central question is deceptively technical: does removing copyright-management information from data used to train an AI model cause a concrete injury sufficient for Article III standing, even without proof the stripped copies were disseminated to the public? Two judges in the same courthouse, considering nearly identical facts filed on the same day, answered differently. McMahon's analysis drew on the Supreme Court's 2021 decision in TransUnion v. Ramirez, which requires plaintiffs alleging a bare statutory violation to show a real-world injury, not merely a technical breach of the statute's text — a doctrine the Second Circuit is now being asked to apply for the first time in the AI training-data context.

A split inside one courthouse

Judge McMahon dismissed the Raw Story and AlterNet case in November 2024, holding the outlets hadn't shown a concrete injury from the CMI removal and so lacked standing to sue — she was openly skeptical an amendment could fix that defect. Judge Rakoff, hearing The Intercept's parallel case, went the other way in a February 2025 written opinion (following a bench ruling the previous November), holding The Intercept plausibly alleged OpenAI intentionally removed copyright-management information and sustaining that claim, while dismissing a related distribution claim and all claims against Microsoft. McMahon then rejected Raw Story and AlterNet's amended complaint in April 2025, again finding no concrete injury, and directed them to the appeals court rather than allow further amendment.

On appeal, Raw Story and AlterNet argue McMahon's dissemination requirement effectively insulates AI training pipelines from CMI liability altogether, since the stripped copies used internally for training were never meant to be shown to the public in their altered form — the same underlying fact pattern Rakoff treated as sufficient for standing in The Intercept's case. How the Second Circuit resolves that tension will likely determine whether CMI claims become a viable tool against AI trainers nationwide, not just within the Second Circuit.

Where it stands

Raw Story and AlterNet's case remains dismissed. They filed their notice of appeal on July 16, 2025, and the Second Circuit heard oral argument on March 18, 2026, on whether McMahon wrongly required proof of dissemination to establish standing. As of July 22, 2026, the appeals court has not ruled, and there is no indication of when a decision will issue. The Intercept's claim, on a separate track before Rakoff, is in discovery, and The Intercept has joined other publishers in the pending sanctions motion against OpenAI filed in the related New York Times MDL in July 2026.

Why it matters for training-data licensing

The CMI-removal theory is the cheapest entry point into AI copyright litigation, because it doesn't require a copyright registration — only proof that identifying metadata was intentionally stripped. That's exactly why the standing question matters so much: if the Second Circuit revives Raw Story's claim, any dataset-cleaning pipeline that strips bylines and copyright notices before training could face standalone DMCA liability, independent of whether the underlying training itself infringes anything.

For suppliers, the practical lesson is to document what metadata a data-preparation pipeline preserves or removes, and why. For buyers, that documentation should be a diligence item on its own, separate from copyright chain-of-title — because on the current split, one federal judge in Manhattan already thinks stripping that metadata alone is enough to get sued.