“Above all don’t sign these contracts.” This blunt advice, issued by Nando de Freitas, Vice President of Microsoft AI and former director of DeepMind, has sent shockwaves through the tech community. A post to X containing his words is directed at Google DeepMind workers struggling with restrictive “noncompete” contracts that prevent employees from working for competing companies for as long as a year. Although the agreements do aim to protect proprietary knowledge in a competitive industry, the implications of the pacts are raising ethical and practical questions.
Google’s controversial leading AI research division, DeepMind, has reportedly adopted a strategy that sends some staff into months of extended “garden leave” when they quit. During this time, staff are paid but forbidden from working for rivals. These agreements also vary depending on seniority and role, with some researchers facing downgraded restrictions of 12 months, according to former employees. For individual contributors on teams behind projects such as Google’s Gemini AI models, six-month noncompetes are standard. But senior researchers have been bound by yearlong clauses, a duration that one former employee said felt like “forever in AI.”
The competitive environment for artificial intelligence has changed dramatically since the launch of OpenAI’s ChatGPT in late 2022. As companies scramble to create paradigm-shifting technologies, keeping the best talent has become a strategic necessity. But the rapid speed of innovation in AI makes long noncompetes especially fraught. Often, startups and competitors like OpenAI and Microsoft which are moving south in a big way want instant hires, or those whom you will have to wait to hire due to noncompetes are played into a disadvantage.
Last year, the United States’ Federal Trade Commission (FTC) outlawed most noncompete agreements, encouraging startups and creating a more mobile labor market. But this ruling has no bearing on noncompetes enforceable at DeepMind’s London headquarters, where it is still possible to enforce a noncompete if a judge believes it is reasonable to the fulfillment of business interest. DeepMind employees have even contemplated moving to the U.S., notably California, where noncompetes cannot be enforced according to state law. California’s 2023 legislation went one step further and expanded this ban, highlighting the divergent American and European approaches to employment contracts.
Industry leaders like de Freitas have publicly condemned the ethical implications of these agreements. In his post, he congratulated DeepMind for its recent accomplishments in AI but referred to the restrictive contracts as an “abuse of power.” He encouraged employees to interact with internal leadership, specifically naming the CTO, Koray Kavukcuoglu, and the Senior Research Director Douglas Eck, both of whom he asserted oppose the policy. Your leads are responsible for this. Talk to them, he wrote, underscoring the importance of dialogue within the organization.
Google has defended its use of noncompetes, noting that its employment contract terms are in line with market standards and that they are applied only selectively to protect sensitive work. But this explanation has not assuaged fears among employees or industry watchers. As generative AI investment continues to pour in, the restrictive clauses are being compared with practices in the financial sector, where hedge funds have long employed similar tactics to hold on to key talent.
These agreements do not just limit job movement for single careers; they have broader impacts. They repress innovation, critics say, by limiting the flow of ideas and know-how between organizations. In an industry that can turn on a dime in six months, the power to pivot is key. “AI is interesting. It seems to be the first time in my career that you have this insane race, like a space race.”
For DeepMind staff, this is made more complicated by workplace culture. Some staff members are told by their managers that getting job offers from other companies is a way to secure a promotion within the company, de Freitas said, a practice he called concerning. His message to employees was clear: “Please don’t reach out to me. Rather reach out to each other.”
With an analysis on new training focused on AI transcript data that are breaking conventional workplace settings, comparisons are drawn to the AI sector’s competitive ecosystem and the need for standards that prioritize innovation and workforce longevity over guarded intellectual property. As companies such as Google and Microsoft compete for market control, the debate over noncompetes will likely grow, a snapshot of wider conflicts in an industry driven by fast change and high stakes.
If you’re in this boat, one word of warning: moving between roles and organizations is not only physically important but it’s also critically important to the future of AI. How practices that prevail now might evolve to cater to the needs of this rapidly advancing field is an open question.
For more on Google DeepMind’s use of noncompetes, or more about the challenges faced by AI professionals, the ongoing conversation presents a glimpse into the bitter realities of this hot new market.
