Why Non-Competes for AI may be too broad in today’s economy

The AI talent war is in full swing. Meta, Google, Microsoft, OpenAI, Anthropic, AWS, Nvidia, Safe Super Intelligence, and Thinking Machines Lab are all fiercely competing for engineers, researchers, and machine learning professionals. Five years ago, non-compete agreements in artificial intelligence were rare and often targeted only a handful of elite research roles. Today, as generative AI and large language models become the backbone of nearly every workflow and product, the reach of non-competes is expanding—sometimes to the point of absurdity.

This article explores why non-competes for AI roles are increasingly viewed as overbroad, how the legal landscape is shifting, and what both employers and employees need to know to navigate this new reality.

The Expanding Scope of “Competition” in AI

In 2019, AI and machine learning were still considered niche specialties. Most non-competes targeted a small pool of researchers at places like DeepMind, OpenAI, or Google Brain. Now, generative AI and large language models are embedded in everything from cloud infrastructure (AWS, Microsoft Azure) to chip design (Nvidia) and even the operations of emerging labs like Safe Super Intelligence and Thinking Machines Lab.

The result? A non-compete that restricts “AI” or “machine learning” work could, in effect, bar someone from working in almost any meaningful tech role. This is similar to what happened in the early 2000s, when non-competes in “internet” or “information technology” became so broad that they were often unenforceable.

Defining “Competition” in the Age of Generative AI

The core challenge is that AI is no longer a siloed field. Large language models and machine learning algorithms are now foundational to products and services across every industry—healthcare, finance, logistics, entertainment, and beyond. If a non-compete prohibits working on “AI,” it could prevent an engineer or researcher from joining any company that uses AI in any capacity, which is nearly all of them.

For example:

• Meta, Google, Microsoft, OpenAI, Anthropic, AWS: All have teams working on large language models, generative AI, and applied machine learning.

• Nvidia: Its chips power nearly every major AI system, making its engineers and researchers central to the entire ecosystem.

• Safe Super Intelligence and Thinking Machines Lab: These labs are pushing the boundaries of what AI can do, but their work is quickly adopted by mainstream tech companies.

The line between “competitor” and “non-competitor” is now so blurred that broad non-competes risk being both unfair and unenforceable.

Legal and Practical Limits

State Law Trends

• California: Non-competes are void and unenforceable. This has made the state a magnet for AI talent and innovation.

• Massachusetts: Non-competes are only enforceable for employees above a certain income threshold, and employers must provide “garden leave” (paid time off during the restriction).

• Washington: Requires advance notice and limits duration and scope.

• New York, Texas: Non-competes are still generally enforceable if reasonable, but courts are increasingly skeptical of broad restrictions—especially in fast-moving fields like AI.

Recent Court Cases

Courts are striking down non-competes that are too broad in scope, duration, or geography. In several recent cases, judges have found that restrictions covering all “AI” or “machine learning” work are unreasonable, especially as these technologies become ubiquitous.

Practical Realities

• Remote Work: With distributed teams, geographic restrictions are harder to justify.

• Rapid Innovation: In AI, a year out of the field can mean missing an entire generation of technology. Long non-competes are especially damaging.

Industry Examples

• Meta, Google, Microsoft, OpenAI, Anthropic, AWS: These companies have shifted away from broad non-competes, focusing instead on protecting specific trade secrets and using NDAs.

• Nvidia: Non-competes are narrowly tailored to protect proprietary chip designs or confidential architecture, not all “AI work.”

• Safe Super Intelligence, Thinking Machines Lab: These labs use creative retention strategies—like equity vesting and strong NDAs—rather than broad non-competes.

Alternatives to Non-Competes

Given the legal and practical challenges, many companies are turning to alternatives:

• Non-Disclosure Agreements (NDAs): Protect confidential information without restricting future employment.

• Non-Solicitation Clauses: Prevent departing employees from poaching clients or colleagues, rather than barring them from the entire field.

• Trade Secret Law: Offers robust protection for proprietary algorithms and data, without the downsides of non-competes.

Thinking Machines Lab, for example, relies on a combination of NDAs and internal culture to protect its innovations, rather than trying to lock down talent with unenforceable contracts.

Why Narrowly Tailored Agreements Are the Only Sustainable Path

The legal trend is clear: non-competes that attempt to cover all “AI” or “machine learning” work are increasingly seen as unreasonable and unenforceable. Courts, regulators, and the industry itself are moving toward more targeted restrictions—if any at all. For Meta, Google, Microsoft, OpenAI, Anthropic, AWS, Nvidia, Safe Super Intelligence, and Thinking Machines Lab, the focus is shifting to protecting true trade secrets and fostering innovation, not stifling talent mobility.

Conclusion

As generative AI and large language models become foundational to every industry, non-competes that attempt to cover all “AI work” are simply too broad for today’s economy. The future belongs to companies that can protect their core innovations while allowing engineers, researchers, and machine learning professionals to move freely and contribute to the broader ecosystem. In this new era, narrowly tailored agreements—and a culture of trust—are the only sustainable way forward.