Written by: Wiktor Stefański, Head of People & Operations, Digital Colliers
The USA Today lawsuit is one of dozens now working through federal court. Publishers are suing OpenAI, Anthropic, and Meta. The claims follow a pattern: the model vendor scraped and copied copyrighted articles to train its systems without permission or payment. Counsel advising media or publishing clients will field questions they cannot answer yet. Which vendors used our content? What evidence exists? What remedies can we pursue? Most firms do not have a workflow to answer those questions quickly.
The pattern emerging in depositions and discovery
The publisher lawsuits share a common structure. Plaintiffs produce server logs, robots.txt files, and content indexes. They show the defendant's crawlers hit their sites millions of times. Discovery requests demand training-data inventories, model weights, and internal correspondence about content sourcing. Some vendors disclose partial training sets. Others invoke trade-secret protection and move to limit discovery. The publishers argue that without full disclosure, they cannot prove which articles ended up in which model. Courts are still working out how much access plaintiffs get to proprietary training pipelines. But every new filing tightens the precedent. Counsel who wait for bright-line rules will be drafting motions on short notice when their client gets served.
Why in-house teams cannot wait for the case law
Your publishing client will ask the questions as soon as they read the headlines. Did OpenAI train on our archive? What about Anthropic, Google, Meta? Do we have licensing agreements that allowed it? If not, what is the damages calculation? In-house teams need answers in days, not months. They need to know whether to send a preservation letter, whether to join an existing lawsuit, or whether a licensing deal makes more sense. You cannot build that analysis during the first call. You need a repeatable workflow that runs before the phone rings. The EU AI Act adds regulatory pressure. Article 50 transparency obligations apply from August 2026. Vendors deploying foundation models in Europe must disclose training-data sources in technical documentation. Fines reach up to €15M or 3% of global turnover. If your client operates in both US and EU markets, you need disclosure workflows that satisfy both regimes.
What the four-step workflow looks like
The workflow has four stages. First, build a content inventory. Catalogue everything your client publishes: articles, images, video transcripts, paywalled archives. Note publication dates, copyright registrations, and any Creative Commons or syndication licenses already in place. Most publishers have partial inventories in their CMS. You need a unified view that maps content to IP ownership and licensing status. Second, map the vendor landscape. Identify which model vendors operate in your client's markets. OpenAI, Anthropic, Google, Meta, Mistral, Cohere, and others. Track their public statements about training data. Some vendors publish data cards or model cards that list sources. Others stay silent. Build a matrix: vendor, model name, disclosed training sources, deployment jurisdictions. Third, check existing licenses. Many publishers signed syndication or API deals years ago. Some agreements included broad grant-back clauses that might cover AI training. Others were narrowly scoped to search indexing or news aggregation. Pull every agreement that touches content distribution. If the language is ambiguous, flag it for renegotiation. Fourth, plan your disclosure timeline. If you are advising the plaintiff, you need to preserve evidence and draft discovery requests. If you are advising a vendor, you need to prepare privilege logs and trade-secret assertions. Either way, the timeline compresses once litigation starts. Courts are moving faster on AI cases than they did on earlier technology disputes. One tracker counted court cases involving AI-fabricated citations rising from 87 to over 1,300 in eleven months during 2024. Judges are learning quickly. They will not grant indefinite extensions while you build your workflow from scratch.
What operators shipping this workflow do differently
Firms that have this workflow running treat it like regulatory compliance, not bet-the-company litigation prep. They update the content inventory quarterly. They monitor vendor announcements and model releases monthly. They review licensing agreements annually, not only when a lawsuit hits. They train associates to spot ambiguous grant-back clauses during contract review. The workflow becomes part of standard due diligence for any client in media, academic publishing, or content production. It is not expensive to maintain once you build it. But it is expensive to build under deposition deadlines. The USA Today lawsuit will not be the last. The next dozen are already being drafted. Counsel who build the workflow now will answer their clients' questions in hours. Everyone else will be explaining why they need three weeks and outside consultants.