OpenAI Fights New York Times Request for 20 Million ChatGPT Conversations in Major Privacy Battle

The Discovery Weapon: How a Copyright Case Became Silicon Valley’s Privacy Precedent

When Legal Discovery Turns Into Data Surveillance

OpenAI’s declaration on November 12 marked a turning point in the battle between artificial intelligence companies and traditional media. By rejecting The New York Times’ sweeping discovery demands, OpenAI didn’t just push back on a lawsuit—it drew a line between legal discovery and large-scale data surveillance.

The Times isn’t asking for internal memos or training data. It wants something far more personal: 20 million ChatGPT conversations from December 2022 through November 2024. Imagine that—an enormous sample of human thoughts, secrets, and daily chatter poured into an AI interface. The newspaper’s goal is to find proof that users used ChatGPT to dodge its paywall, claiming the AI indirectly competes with and erodes journalism’s last stable revenue stream.

But this fight cuts much deeper than a single lawsuit. If the court sides with the Times, future AI lawsuits could turn into privacy invasions on a massive scale. Plaintiffs might demand access to millions of user conversations, turning discovery into a fishing expedition across people’s private digital lives. OpenAI originally faced a demand for 1.4 billion chats before bargaining it down to 20 million—a “compromise” still bigger than most criminal investigations in U.S. history.

The legal details matter here. In May 2025, U.S. Magistrate Judge Ona T. Wang ordered OpenAI to preserve all conversation logs indefinitely, overriding its standard 30-day deletion rule and impacting more than 400 million users. Although parts of that order were later relaxed for operational data, a historical sample remains locked under court hold. OpenAI insists this is about protecting user privacy, while the Times accuses the company of obstruction—especially since another AI firm already handed over 5 million user chats in a different case.

The Asymmetry Problem in Discovery Proportionality

This legal tug-of-war exposes a deeper issue: scale. AI doesn’t operate in the same realm as traditional lawsuits. When OpenAI offered privacy-preserving alternatives—like keyword searches for Times content, categorized summaries, and strict de-identification protocols—the Times turned them down.

Why? Because its lawsuit relies on two claims that need different types of evidence. The first says OpenAI illegally trained ChatGPT on Times articles, which involves training data. The second argues that ChatGPT helps users bypass the Times’ paywall and regurgitate its content—something provable only through large-scale user behavior data. That’s why the newspaper insists on reviewing those 20 million conversations.

Here’s where things get messy. Most of those chats have nothing to do with the Times at all. They include therapy-like confessions, personal finance advice, health questions—private moments between users and a tool they thought was secure. The Times wouldn’t view them directly; instead, outside lawyers and technical contractors would handle them under protective orders. Privacy advocates warn that even this setup is far too porous for data so personal.

Discovery rules were meant to prevent fishing expeditions, but those old standards are buckling under AI’s data volumes. A normal lawsuit might involve thousands of documents. This one involves tens of millions of conversational snippets, each a potential minefield of personal details that algorithms can’t always scrub clean.

Wall Street’s Discovery Tax: Why Markets Are Watching Court Procedure

Investors are paying close attention. This isn’t just about copyright anymore—it’s about how discovery itself could reshape AI’s business model. Three big financial pressure points stand out.

First, there’s discovery as a recurring cost. If 20-million-chat productions become the new norm for AI lawsuits—and with more than 50 similar cases pending—that turns one-off legal costs into permanent overhead. Companies will need to build full-time systems for data scrubbing, encryption, and secure review. These costs rise with user base size. For OpenAI’s roughly 800 million weekly ChatGPT users, that’s not pocket change—it’s a structural shift. Smaller AI labs may not survive that burden.

Second, there’s encryption versus innovation speed. OpenAI’s promise to add client-side encryption and automated privacy controls will limit how much real user data can be used for model improvement. If encryption becomes standard, model quality will depend more on expensive licensed data and opt-in corporate sources. That gives giants like Microsoft, Google, and Apple an edge—they can afford to buy the data. OpenAI’s real goal here seems clear: remove itself from the data-custody chain altogether.

Third, media leverage is shifting. The Times’ aggressive push for discovery, even if trimmed later, strengthens publishers’ hands in negotiations with AI companies. Every major outlet can now cite this case as precedent when demanding licensing fees. Suddenly, the threat of discovery becomes a bargaining chip. Deals like News Corp’s and Axel Springer’s look brilliant in hindsight.

On Wall Street, The New York Times Co. trades near $63, close to record highs. Investors see stability in subscriptions and upside in AI litigation. But there’s a PR cost here too—being painted as the company demanding access to millions of strangers’ private chats could sour public opinion and complicate a jury’s view. Meanwhile, Microsoft’s staggering $3.85 trillion valuation barely flinches; to investors, this fight is background noise. Still, faster, enterprise-grade privacy tools born from this dispute could give Microsoft’s Copilot a big trust advantage.

Most analysts expect the court to land somewhere in the middle. The likeliest outcome? A narrowed review, limited to chats that actually match Times content, examined under strict conditions. That would test the paywall-bypass claim without normalizing mass surveillance. Odds of that happening: roughly 60–70 percent within the next six months.

The Institutional Learning Curve

Here’s the real takeaway: the legal system is learning, in real time, that old discovery rules don’t fit the new AI world. Twenty million conversations aren’t just “documents.” They’re reflections of thought, emotion, curiosity, and identity—packaged as data but fundamentally human. Courts simply aren’t built to handle that yet.

If OpenAI delivers on its encryption promise in 2026, this entire problem disappears for future cases. After all, you can’t subpoena what no longer exists in readable form. That’s the long game—shut off the data tap, not just win a single lawsuit. Every major AI company is racing to do the same.

This case isn’t really about copyright or fair use anymore. It’s about whether discovery rules designed for paper files can scale to trillion-parameter models trained on the human experience itself. The Times wants to prove ChatGPT steals journalism. But in doing so, it might break something even more fundamental—the quiet assumption that when you talk to a machine, that conversation stays private.

The Convenient Timing of Righteous Indignation

Both parties deserve scrutiny for selective outrage. The Times' 20-million-chat demand isn't new—it's been winding through discovery since summer 2025, with OpenAI proposing this exact sample size in August. OpenAI's November 12 privacy crusade conveniently arrived right after a German court ruled the company violated IP rights. The PR offensive reads less like genuine privacy principle than tactical deflection—weaponizing user sympathy to bury headlines about systematic copyright infringement. Meanwhile, the Times rejected privacy-preserving search alternatives months ago but only now faces public backlash for it, benefiting from OpenAI's news-cycle manipulation. The discovery fight is real; the timing of this public battle is theater designed to reshape narratives when both sides need cover from their own contradictions.

NOT INVESTMENT ADVICE