Frontier Security Institute
Last reviewed
Jun 3, 2026
Sources
4 citations
Review status
Source-backed
Revision
v1 · 1,242 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 3, 2026
Sources
4 citations
Review status
Source-backed
Revision
v1 · 1,242 words
Add missing citations, update stale details, or suggest a clearer explanation.
The Frontier Security Institute (FSI) is a Washington, D.C. organization launched by the Center for AI Safety (CAIS) to connect frontier artificial intelligence developers with the United States national security establishment. CAIS announced the institute on June 2, 2026, describing it as a "translation layer" between the private labs building advanced AI and the government bodies, the Pentagon, the intelligence community, Congress, and allied institutions, that have to acquire, govern, and operate those systems.[1][2] The launch came alongside a leadership shake-up at CAIS, which named former xAI engineer Devin Kim as its first president.[1]
The institute is led by Isaac "Ike" Harris, a Navy veteran, as executive director.[1][3] Its creation reflects a wider shift in which AI safety groups, once focused mainly on technical research and public advocacy, have begun building dedicated channels into defense and intelligence agencies.
CAIS is a nonprofit research organization based in San Francisco. It was founded in 2022 by the machine learning researcher Dan Hendrycks and Oliver Zhang, and works on reducing what it calls societal-scale risks from AI through technical research, field-building, and policy advocacy.[4] The group is probably best known for a one-sentence statement it published in May 2023 warning that "mitigating the risk of extinction from AI should be a global priority," which was signed by hundreds of researchers and the heads of several major AI companies.[4]
The Frontier Security Institute operates as a CAIS initiative rather than an independent entity, and the June 2026 announcement framed it as part of a broader expansion of the parent group's leadership and reach.[1] Hendrycks, who continues as executive director of CAIS, said the day "marks a significant moment for CAIS and for the field."[3]
FSI's stated goal is to close what CAIS calls the capability gap between frontier AI and the national security enterprise. The argument runs like this: unlike past strategic technologies, the U.S. government does not hold a monopoly on AI. Frontier capability is built in private labs, on private compute, by private workforces, and it moves faster than any acquisition, oversight, or doctrine cycle can absorb on its own.[1][2] Without a working channel between the labs producing the capability and the institutions responsible for fielding it, the institute argues, every new model release widens the gap. The cost, in its framing, shows up as military doctrine that lags the technology, acquisition pathways that stall, and openings for adversaries to co-opt American innovation.[1]
The "translation layer" idea is the institute's answer to that gap. Rather than building models or writing regulation, FSI positions itself in between: it aims to take what frontier developers understand about their systems and render it usable for the operators, acquisition officials, and policymakers who do not work inside the labs, and to carry the government's requirements and constraints back the other way. The two communities speak different languages, roughly the language of research and the language of operations, and FSI's premise is that someone has to do the interpreting.[2]
The institute has said its initial work will focus on problems specific to the national security use case rather than AI risk in general. Those include securing advanced AI models against theft or compromise, the way operators test and evaluate these systems before relying on them, and how AI is reshaping geopolitical stability.[1][2] The emphasis on a national security audience distinguishes FSI from CAIS's earlier work, which was aimed more at the research community and the general public.
Isaac "Ike" Harris serves as the institute's executive director. According to CAIS, Harris spent 23 years as a U.S. Navy surface warfare officer, a career that included command of the destroyer USS Ramage (DDG-61) and deployments to Europe and the Middle East.[1] After his uniformed service he worked as a policy advisor to the secretary of defense on China and technology security, and as a senior professional staff member on the House Select Committee on the Chinese Communist Party.[1] Before joining FSI he was vice president of government strategy at Exiger, a risk and supply-chain analytics firm.[1] That mix of operational, congressional, and private-sector experience maps closely onto the bridging role the institute describes for itself.
CAIS named several other people to the FSI team. Jeremy Pelter, who spent close to two decades in federal service and served as acting U.S. secretary of commerce, was named chief operating officer.[1] Aaron B. Frank, a computational social scientist and former researcher at the RAND Corporation, became director of research.[1] Susan Malandrino, previously a senior advisor to the president of the International Federation of Red Cross and Red Crescent Societies, was named vice president of communications.[1]
The same June 2, 2026 announcement introduced Devin Kim as the first president of CAIS, a newly created role distinct from Hendrycks's position as executive director.[1] According to CAIS, Kim was an early employee at xAI, where he led post-training tooling and research infrastructure for the company's Grok models, and earlier worked as an engineer at Scale AI on content-understanding and trust-and-safety systems. The organization said he had also collaborated with teams at OpenAI and Meta.[1] (The original report describing him as a former xAI and Scale AI engineer is consistent with this account.)
As president, Kim is expected to lead the organization's research mission, organizational strategy, and its growing engagement across the policy, national security, and AI communities, and to strengthen CAIS's field-building work by supplying technical infrastructure to the wider AI safety research community.[1] CAIS framed the appointment around the idea that Kim had built systems at the frontier of AI development and could translate that work for audiences outside the technical world, the same interpreter role that defines the new institute.[1]
The launch fits a pattern that took shape over 2024 and 2025, as AI safety organizations and policy groups moved from publishing research and warnings toward direct engagement with defense and intelligence agencies. Governments on both sides of the Atlantic stood up dedicated bodies, including the U.S. and U.K. AI safety institutes, to evaluate frontier systems, and a string of think tanks began producing national security playbooks for advanced AI. FSI is a nonprofit version of that turn: an attempt to embed safety and evaluation expertise inside the defense conversation rather than commenting on it from outside.
Whether a self-described translation layer can hold a neutral position is an open question. FSI sits between commercial labs with strong incentives to sell capability and a government apparatus with its own priorities, and several of its leaders come directly from defense and national security backgrounds. Supporters see that fluency as the whole point, since someone has to speak both languages credibly. Skeptics may worry that a safety organization moving this close to the Pentagon risks becoming an advocate for adoption rather than a check on it. The institute had only just been announced as of mid-2026, so its actual output, and how independent it proves to be, remained to be seen.