Figma Faces Class Action Claiming It Secretly Trained AI on Millions of Customer Designs Without Permission

Figma's $16 Billion Question: When Customer Data Becomes AI Fuel

A proposed class action exposes the hidden cost of training AI on enterprise designs—and why every SaaS company should be watching

The lawsuit filed November 21 against Figma in Northern District of California reads like a contract dispute. That's precisely what makes it dangerous.

Plaintiff Raza Khan, a startup founder, doesn't primarily allege copyright infringement—the now-familiar battleground where OpenAI, Meta, and Midjourney have sparred with creators. Instead, the complaint frames Figma's alleged use of "millions" of customer design files as breach of contract and misappropriation of trade secrets: you promised our work was private, then fed it to your AI anyway.

The distinction matters. Copyright cases pit fair use against infringement, a legally murky dance. But when plaintiffs can point to specific marketing promises—as the complaint reportedly does, citing years of Figma assurances that customer files "belonged to them" and wouldn't benefit Figma's own product development—courts become far less sympathetic. Add trade secrets into the mix, where startup IP and agency work under NDA allegedly entered Figma's training corpus, and you've constructed a narrative courts have repeatedly validated: the bait-and-switch.

Figma disputes everything. A spokesperson told Reuters the company uses customer data for training only with "explicit authorization," de-identifies it, and focuses on "general patterns—not customers' unique content." Yet the 2024 "Make Designs" controversy, where Figma pulled an AI feature after it generated near-replicas of Apple's Weather app, already seeded doubt about the company's AI guardrails.

The Industry Context: Exception or Trend?

This is no outlier. Since 2023, over 50 major AI-training lawsuits have been filed, with class action exposure tripling by 2025. The pattern is consistent: companies scrape or repurpose user-generated content without clear consent, claim transformative use, then face coordinated legal assault.

Anthropic's $1.5 billion settlement with authors in September 2025—including court-ordered destruction of pirated training data—established a crucial precedent. Courts will mandate algorithmic disgorgement, erasing model weights built on tainted corpora. That outcome terrifies SaaS platforms more than damages: it forces expensive retraining and potentially cripples AI features mid-deployment.

LinkedIn faced a January 2025 class action over sharing private InMail messages for AI training via retroactive policy changes. Google's Gemini team battled similar claims after auto-opting Gmail and Docs users into training. Salesforce is defending against allegations it used "thousands of pirated books" for its XGen models. The common thread? Enterprise or consumer platforms treating first-party data as "free fuel" for AI pivots—then discovering consent wasn't implied.

What separates these from pure content-scraping cases (New York Times v. OpenAI) is the reliance framework: users uploaded sensitive work because platforms promised protection, creating implied contracts that AI training allegedly breached. That's harder to dismiss than fair use debates.

The Investment Calculus: Red Flag, Not Existential Threat

For investors, this case poses governance risk and moderate legal overhang—not a death sentence. Here's the math.

Figma trades at roughly $16-17 billion market cap after its July 2025 IPO raised $1.2 billion. The company is profitable, generating $749 million in 2024 revenue with 88% gross margins. A settlement in the $200-600 million range, while painful, wouldn't crater the business. Even Anthropic's $1.5 billion payout, though staggering, reflects training on a massive external corpus; Figma's exposure may be narrower if enterprise consent mechanisms hold up in discovery.

The real risk isn't the check—it's injunctive relief. If courts determine Figma's AI features rely materially on data obtained through broken promises, they could order model destruction or usage bans. Figma's S-1 mentions AI over 150 times; it's core to the growth story and valuation premium. A forced "AI reset" would compress multiples even if revenue held steady.

Probability-weighted, this translates to:

• Direct settlement risk: $200-600 million (base case), with $1 billion+ as tail scenario if broad class certifies and trade secret theories resonate
• Indirect costs: 2-3 points of operating margin drag over three years from legal fees, compliance buildout, and model retraining
• Strategic drag: Slower AI roadmap deployment, potential churn among NDA-sensitive enterprise customers

Investors should mark a modest multiple discount versus "AI-clean" SaaS peers until Figma demonstrates rebuilt data governance—public opt-in flows, NDA-safe modes, external audits. Watch enterprise customer commentary and net dollar retention for early signals of trust erosion.

The Systematic Risk No One Priced

Figma's real significance isn't company-specific. This case clarifies that "AI trained on first-party SaaS data" is now a regulated asset class, subject to consent requirements courts and regulators are converging around: ex-ante disclosure, real opt-out mechanisms, and unwinding capability for tainted corpora.

That threatens any collaborative platform pitching AI that "learns from your data"—Notion, Canva, Miro, Microsoft 365's Copilot, Salesforce Einstein. The market underpriced how quickly data provenance would become diligence-critical. Figma just made it standard.

NOT INVESTMENT ADVICE