The AI industry’s balance sheet has a line item most capex charts don’t show: legal exposure. A live U.S. lawsuit tracker updated September 24 now counts 142 active AI-related cases, with OpenAI (36) and Anthropic (28) topping the list — a reminder that while the market obsesses over GPU spend and leaderboard rankings, the courts are quietly building a parallel dataset on AI risk. This is a living story, not a one-off rumor roundup, and we’ll keep updating the numbers as the tracker does.
Why it matters
Every dollar of AI infra spend — the hundreds of billions flowing into GPUs, data centers, and cloud contracts — is being deployed by companies that are simultaneously accumulating legal liability at a measurable, tracked rate. Litigation doesn’t show up on a capex slide, but it is a real cost center: settlements, licensing deals, discovery, and injunctions all eat into the margin that infrastructure spend is supposed to protect. Investors modeling AI capex vs. revenue should treat the lawsuit count as a leading indicator of hidden opex, not legal-department noise.
The numbers, as of September 24
- 142 total AI lawsuits tracked in the U.S.
- 92 are intellectual-property claims — nearly two-thirds of the docket, confirming that copyright and training-data disputes remain the dominant legal risk category for model builders.
- 10 are hiring-discrimination cases, tied to AI-driven HR and recruiting tools.
- 6 are product-liability or wrongful-death cases — a small but consequential category given the severity of outcomes at stake.
- New filings in mid-September added cases against Suno, Anthropic, and OpenAI, showing the docket is still growing, not plateauing.
Most-sued companies: a leaderboard of its own
| Company | Active Cases | Primary exposure |
|---|---|---|
| OpenAI | 36 | Copyright/IP, product liability |
| Anthropic | 28 | Copyright/IP |
| Suno | 13 | Music/training-data copyright |
| Perplexity | 10 | Content licensing, IP |
| Workday | 9 | Hiring discrimination |
The pattern is telling: the companies spending most aggressively on frontier model training and inference infrastructure are also the ones absorbing the most legal exposure. That’s not necessarily causal, but it correlates — the biggest capex spenders are the biggest litigation targets, largely because scale means more training data, more outputs, and more surface area for claims.
Court outcomes remain uneven
Volume doesn’t equal victory for plaintiffs. On September 16, the Ninth Circuit affirmed dismissal of a Copilot-related DMCA claim, a reminder that plenty of these 142 cases will fail or narrow rather than produce blockbuster payouts. Readers should resist the instinct to treat “142 lawsuits” as “142 looming liabilities” — the tracker counts filings and active dockets, not verdicts. An honest read of this data set requires holding both facts at once: litigation volume is rising, and courts are still just as often ruling for AI companies as against them.
Builder’s take: what this means if you’re shipping
If you’re building on top of foundation models, the litigation tracker is now as relevant to your risk assessment as cloud GPU economics. Consider three practical angles:
- Vendor concentration risk — relying heavily on a single top-5 sued provider means inheriting some of their legal uncertainty, particularly around training-data provenance.
- Licensing costs may rise faster than compute costs — as IP cases resolve or settle, expect API pricing or licensing terms to shift in ways that don’t track GPU cost curves at all.
- Diversify exposure like you diversify infra — just as teams hedge against cloud GPU price swings by multi-sourcing compute, there’s a case for tracking legal exposure across the vendors in your stack the same way you’d track a leaderboard.
Leaderboards and litigation: two forms of volatility
It’s worth noting the coincidence in timing: while the lawsuit tracker logs a new peak, separate reporting says at least one AI leaderboard was rewritten three times in a single week as scores and cost-efficiency metrics shifted. Both data points tell the same underlying story — the AI market has no stable ranking system yet, whether you’re measuring model quality or legal risk. For anyone modeling AI capex against revenue, that instability cuts both ways: today’s top performer on a benchmark could be tomorrow’s top defendant.
What we’re watching next
This tracker updates continuously, and we’ll revisit the count, the top-5 sued list, and notable rulings as they land. The next inflection points to watch: whether any of the 92 IP cases produce a landmark ruling on fair use for training data, and whether hiring-discrimination cases against Workday-style vendors expand beyond the current 10.