Luke Metz, a machine learning researcher who left OpenAI for Thinking Machines Lab (TML) and returned to OpenAI within the same year, has resigned again — this time to join Meta Superintelligence Labs (MSL), the research unit Mark Zuckerberg built around Alexandr Wang after acquiring a stake in Scale AI. According to Techmeme, the move makes Metz one of a growing list of researchers who have cycled between OpenAI, independent labs, and Meta inside a single calendar year.

The specifics of what Metz will work on at MSL have not been disclosed. What's notable is the pattern: a researcher senior enough to be recruited by Thinking Machines Lab, then re-recruited back by OpenAI, is now being pulled again — this time by Wang's operation inside Meta. Three employers, one person, roughly twelve months.

For an outside observer, one researcher switching jobs is not news. What makes it worth tracking is what it says about the stability — or lack of it — inside the labs building frontier models.

A revolving door, not an outlier

Metz's trajectory mirrors a broader churn that has defined 2025 and now 2026 for the top AI labs. OpenAI has seen researchers depart for and return from Anthropic, Thinking Machines Lab, and Meta in overlapping waves; Meta, in turn, has spent heavily to stand up MSL as a credible fourth pole alongside OpenAI, Google DeepMind, and Anthropic. Wang, who joined Meta after it took a large stake in Scale AI, has been the public face of that recruitment push.

• Researchers moving between labs bring more than code — they bring context on training runs, evaluation methodology, and unpublished failure modes.
• Repeated departures and returns (as with Metz's OpenAI–TML–OpenAI–Meta sequence) suggest compensation and mission alignment are being renegotiated constantly, not settled once.
• Non-competes are effectively unenforceable in this labor market — the practical constraint on a researcher's mobility is an NDA and a vesting schedule, not a legal wall.

Why Meta is the one attracting the trade

Meta's pitch under Wang has been aggressive by design: MSL was built specifically to close the gap with OpenAI and Google on frontier model capability, and that mandate has translated into visible poaching, not just organic hiring. Meta doesn't need Metz to be a household name to benefit from the hire — what it needs is depth on the research staff who have already worked inside OpenAI's training and evaluation pipeline, since that experience compounds faster than starting from scratch.

The counterpoint is that OpenAI has absorbed this kind of churn before without visible damage to its release cadence. A single senior researcher leaving rarely shows up in product roadmaps within the same quarter. The risk is cumulative, not immediate: if MSL keeps winning these trades one hire at a time, the gap it's trying to close narrows one person at a time too.

What this means if you build on top of these labs

For teams building products on OpenAI's or Meta's APIs, individual researcher moves are not directly actionable — but the pattern is worth watching for a few practical reasons:

• Roadmap signal. A lab that is actively losing and re-recruiting research staff around a specific area (in Metz's case, the details aren't public) may see near-term shifts in what gets prioritized for release.
• Model behavior drift. Research staff turnover can precede changes in how a lab tunes its models — new hires often bring their own priors on alignment, evaluation, and tradeoffs.
• Vendor concentration risk. If your product depends on one lab's API, sustained talent churn at that lab is a slow-moving signal worth tracking alongside uptime and pricing, not a reason to panic on its own.

AiiN's takeaway

The battle for talent between AI labs is, in our estimation, currently a sharper contest than the battle between their models — benchmark scores move in small increments release to release, while a researcher's employment status can flip overnight and take non-public methodology with it. Metz's move is a small data point, but it's a clean one: it shows that even researchers who have already tried leaving OpenAI once, and come back, are still willing to leave again when a well-funded challenger like MSL comes calling. Expect more of these moves to surface on Techmeme through the rest of 2026 as Meta continues to staff up MSL and the other labs respond in kind.