# Dan Hendrycks Machine learning researcher and director of the **Center for [[AI Safety]] (CAIS)**, and one of the most consequential connectors between the AI-safety movement and the national-security establishment. Hendrycks's trajectory — from benchmark researcher to co-author of grand strategy with [[Eric Schmidt]] — tracks the AI-safety field's broader migration from existential-risk philosophy toward Pentagon-adjacent "competitiveness" politics. ## Research Credentials Hendrycks earned his PhD at UC Berkeley and made foundational technical contributions before turning to safety advocacy: - **GELU** — Co-created the Gaussian Error Linear Unit, the activation function used in most modern transformer models, including the GPT series. - **MMLU** — Created the Measuring Massive Multitask Language Understanding benchmark, for years the standard measure of LLM capability. - **Humanity's Last Exam** — Co-developed with [[Scale AI]]: a frontier benchmark designed to be the hardest expert-level test for AI models, created as models saturated MMLU. - **Textbook** — Published *Introduction to AI Safety, Ethics, and Society* (2024). His benchmarks are a quiet form of power: the people who define how AI capability is measured shape how progress — and danger — gets perceived. ## Center for AI Safety CAIS is the San Francisco nonprofit behind the **May 2023 one-sentence statement** — "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war" — signed by [[Sam Altman]], [[Demis Hassabis]], [[Dario Amodei]], and hundreds of researchers. It put extinction risk into mainstream political discourse in a single news cycle. **June 2, 2026:** CAIS announced a major expansion — Devin Kim appointed president, and the launch of the **Frontier Security Institute (FSI)**, a Washington, D.C. organization explicitly framed as "the translation layer between frontier AI development and the National Security Enterprise." The safety movement's institutional drift toward the natsec world, made formal. ## The Industry Positions Hendrycks advises two of the most strategically positioned AI companies simultaneously: - **[[xAI]]** — Safety adviser to [[Elon Musk]]'s lab since its 2023 founding. - **[[Scale AI]]** — Adviser since November 2024, alongside the Humanity's Last Exam collaboration with [[Alexandr Wang]]. He takes a symbolic $1 salary and no equity from both — a structure designed to neutralize conflict-of-interest critiques. The influence is the point, not the compensation: he is the safety conscience embedded in both Musk's and Wang's operations, two firms deeply enmeshed in Pentagon AI work. ## Superintelligence Strategy and the Schmidt Orbit In **March 2025**, Hendrycks co-authored **"Superintelligence Strategy"** with [[Eric Schmidt]] and [[Alexandr Wang]] — the paper proposing **MAIM (Mutual Assured AI Malfunction)**, a deterrence regime modeled on nuclear MAD in which any state's bid for unilateral AI dominance invites preventive sabotage by rivals. The paper argues against a Manhattan Project for AGI in favor of deterrence, nonproliferation, and competitiveness. Full treatment on [[Eric Schmidt]]. Hendrycks is also an **AI2050 fellow** — a fellowship funded by [[Schmidt Sciences]], Schmidt's philanthropy. The safety researcher warning about superintelligence is, in part, funded by the investor whose portfolio profits from the race he's warning about (see [[Hillspire]]). ## RWKB Significance Hendrycks is the clearest individual embodiment of the AI-safety movement's bargain with power: technical credibility and extinction-risk framing exchanged for seats at the natsec table. Whether that bargain constrains the race or legitimizes it is the open question — MAIM and the FSI both treat great-power AI competition as a given to be managed, not a choice to be contested. That framing is precisely the one [[Eric Schmidt]]'s commercial empire needs to be true. See also [[TESCREAL]] for the ideological waters this swims in. ## See Also - [[Eric Schmidt]] — Superintelligence Strategy co-author; AI2050 funder - [[Alexandr Wang]] — Superintelligence Strategy co-author; Scale AI - [[Elon Musk]] — xAI founder - [[Scale AI]] - [[xAI]] - [[TESCREAL]]