All of xAI's co-founders are gone. That is a procurement question, not a soap opera.

According to an account published by cryptobriefing.com and credited to twinfm.com, xAI had lost six of its original 12 co-founders by late February 2026, and by late March every one of the 11 original co-founders had departed, alongside more than 80 researchers and engineers ^[1][2][3]. The interesting part is not the gossip; it is that enterprise buyers are signing multi-year commitments to model platforms whose founding technical benches are emptying faster than the contracts amortise.

Start with the sourcing, because it matters. The same article says 12 original co-founders in one sentence and 11 in the next, which means at least one of the two numbers is wrong and the whole item is a single chain of attribution rather than an independently confirmed count ^[12][13]. Taken at the lower figure, five co-founders left inside roughly one month ^[14]. Named departures include Tony Wu and Jimmy Ba, both described as part of the original founding team ^[5].

The pattern is not confined to one company. Google's chief scientist Jeff Dean left in February 2026 after nearly 27 years ^[4], at a company that the same source notes invented the transformer architecture underlying virtually every modern large language model ^[8]. At OpenAI and Anthropic, researchers reportedly raised alarms in February 2026 about strategic direction, specifically advertising integration and a perceived deprioritisation of safety research ^[6]. Anthropic's whole market position was built as the safety-first alternative to OpenAI ^[7], so internal dissent on that point is a product claim under stress, not just an HR matter. None of this is new as a labour story: the 2025 State of Tech Talent Report already documented a competitive talent war driven by exactly these exits ^[9].

For an operator, the consequence is continuity, not sentiment. Model roadmaps, fine-tuning support, safety tooling and the undocumented judgement about why a system behaves the way it does live in specific people. When the founding cohort of a vendor turns over inside a quarter, the roadmap you were shown in the pitch was authored by people who are no longer there to deliver it, and the migration cost of a swapped-out model family lands on you, not on them. The cryptobriefing piece frames this as a repricing risk for public equities such as Alphabet if the drain continues ^[10], and as an advantage for well-funded startups that can buy credibility and institutional knowledge by hiring senior researchers out of xAI or OpenAI ^[11]. Both framings point the same way for a buyer: the technical depth you are paying for is mobile, and it may be moving to a vendor you have not diligenced.

The practical asks are unglamorous. Ask who owns your model roadmap by name and what happens if they leave. Ask for deprecation notice periods measured in quarters rather than weeks, written exit assistance, and portability of prompts, evaluations and fine-tuning artefacts. Price a second-vendor fallback into the business case rather than treating it as paranoia.

Worth watching: whether any of the counts in this account get confirmed by a primary source, whether xAI names replacements with comparable publication records, and whether the OpenAI and Anthropic safety complaints become resignations. Also watch where the leavers land, because the startup that hires eight of them becomes a credible vendor and a credible single point of failure at the same time.