CAIS Names xAI Veteran Devin Kim President, Creates Security Unit

The Center for AI Safety is making a structural bet that artificial intelligence safety needs to operate closer to power. On June 2, the San Francisco nonprofit announced Devin Kim as its new President and established the Frontier Security Institute, a Washington, D.C.-based organization built to bridge frontier AI development and the U.S. National Security Enterprise.

Kim arrives from xAI, where as an early employee he built post-training tooling and research infrastructure for the Grok model family. Before that, he worked at Scale AI on content understanding and trust and safety systems, a combination of roles that put him at the intersection of model-building and model governance long before most safety researchers had direct access to frontier infrastructure.

The Washington play

The Frontier Security Institute will be led by Isaac "Ike" Harris, a 23-year U.S. Navy veteran who previously served as a senior advisor to the Secretary of Defense on China technology security. Harris brings institutional credibility that technical expertise alone cannot replicate in those rooms. Framing FSI as a translation layer between AI labs and the national security apparatus is precise language: it acknowledges a real gap, not just in technical knowledge but in trust, vocabulary and authorization structures.

Dan Hendrycks, CAIS's Executive Director, framed both moves as responses to urgency. "Frontier AI is now a national security technology," he said, "and the National Security Enterprise needs partners fluent in both worlds." According to the Yahoo Finance announcement, CAIS describes the dual announcement as a major expansion of leadership and reach as the organization grows to meet safety issues it considers increasingly urgent.

Why this hire signals something

Kim's departure from xAI is worth examining on its own terms. Grok has maintained a competitive release cadence throughout 2026, with llm-stats.com recording Grok 4.3 as released in early May. Someone who built post-training infrastructure for that model family carries working knowledge of what frontier systems can and cannot do that few safety researchers share. His arrival at CAIS suggests the organization is actively recruiting from the production side of the industry, not just the evaluation side.

That production-side perspective matters for an organization trying to establish policy influence. The AI Release Tracker covers more than 160 frontier models released since late 2022, a pace that has consistently outrun the capacity of safety research to keep up. CAIS is betting that engineers who have shipped frontier systems can assess their risks more precisely than those who study outputs in isolation.

On context and implications

Harris's selection as FSI director points to a specific strategic judgment. China technology competition has become the dominant frame through which the National Security Enterprise evaluates artificial intelligence risk, and Harris's advisory role on exactly that topic gives FSI a credentialed entry point into those conversations. Safety research that only speaks to capability benchmarks will be peripheral in that environment. Safety research backed by a 23-year Navy career will not.

For practitioners, what remains unclear is whether FSI will produce technically substantive output. Humanity Redefined has chronicled how rapidly the frontier model ecosystem is moving, with major releases from multiple labs arriving in rapid succession. FSI's real test will be whether it can evaluate those systems rigorously enough to be useful to national security officials who have little patience for assessments that cannot be operationalized.

CAIS built its public reputation on rigorous research and the 2023 statement on AI risk that drew signatures from many of the field's leading figures. Whether it can sustain that rigor while simultaneously operating as a Washington policy shop is the central question this dual announcement leaves open.

FAQ

What is the Center for AI Safety?
CAIS is a San Francisco nonprofit focused on reducing large-scale societal risks from AI, known for technical research and for organizing broad expert statements on existential and catastrophic risk.

Who is Devin Kim and what did he do at xAI?
Kim was an early xAI employee who led post-training tooling and research infrastructure for the Grok model family. Before xAI, he worked at Scale AI on content understanding and trust and safety systems.

What is the Frontier Security Institute?
The FSI is a new Washington, D.C.-based organization established by CAIS to serve as a bridge between frontier AI developers and the U.S. National Security Enterprise. It is led by Isaac Harris, a former senior Defense Department advisor on China technology security.

How does FSI differ from CAIS's existing work?
CAIS has historically focused on technical safety research. FSI represents an institutional expansion into national security policy, explicitly targeting the translation gap between AI lab capabilities and government decision-making.