Update August 5, 2026: Accounting for strict garden leaves, the friction of verifying specific 'senior' job titles, and aggressive retention efforts by established labs led to a slight downward revision in the median and upper bounds for the projected headcount.
As of mid-2026, Safe Superintelligence (SSI) operates with an intentionally lean headcount of roughly 50 employees and maintains extreme corporate secrecy finance.yahoo.com. Currently, only two publicly verifiable individuals clearly meet the strict criteria for prior senior technical roles at the big-three labs: co-founder Ilya Sutskever (former OpenAI Chief Scientist) en.wikipedia.org and Daniel Levy (former OpenAI optimization team lead) 2 sources. A third co-founder, Daniel Gross, came from an Apple AI background and has since departed for Meta en.wikipedia.org.
Beyond the founders, SSI’s known hiring composition has skewed heavily toward Israeli academics, algorithmic trading engineers, and researchers from Google Research—which does not qualify as Google DeepMind under the resolution criteria en.globes.co.il. Recent high-profile senior departures from OpenAI's safety and research leadership teams throughout 2025 and 2026 have largely landed at Anthropic or other competitors, rather than SSI businessinsider.com. Furthermore, SSI’s culture of secrecy actively discourages employees from listing their affiliation on public professional profiles like LinkedIn 2 sources, establishing a strong downward bias on any publicly resolvable headcount.
The primary upside risk to this estimate stems from SSI's July 2026 partnership with Nvidia. This arrangement, which includes a reported ~$5B equity investment and an order-of-magnitude increase in compute access, is explicitly framed as accelerating SSI's "next stage of growth" nvidianews.nvidia.com. This influx of capital and compute could easily fund a strategic pivot away from the lean-team model toward a rapid scaling phase over the next 12 months. Armed with competitive packages benched.ai, SSI could begin aggressively poaching experienced infrastructure, reinforcement learning, and pretraining leads directly from OpenAI or DeepMind to train frontier-scale models.
The median estimate of 8 assumes the two confirmed qualifying founders are joined by a modest number of currently hidden or soon-to-be-hired senior alumni, balancing the new Nvidia funding against SSI's established preference for non-marquee talent. The long right tail reflects a scenario where the recent capital injection triggers an aggressive, public lift-out of elite big-three talent. Conversely, the low left tail accounts for the real possibility of stagnation, further attrition, acquisition by a larger tech firm, or strict resolution constraints where extreme secrecy prevents verifying anyone beyond Sutskever and Levy.
Accounting for strict garden leaves, the friction of verifying specific 'senior' job titles, and aggressive retention efforts by established labs led to a slight downward revision in the median and upper bounds for the projected headcount.