Michael Kratsios: Inside the White House's AI Strategy

Overview

Michael Kratsios, Director of the White House Office of Science and Technology Policy (OSTP) and a senior science/technology advisor to President Trump, discusses the Trump administration’s AI and broader science policy approach. He emphasizes regulatory clarity, support for open-source, and avoiding overly rigid rules that lag behind rapidly evolving technology.

Key points and arguments

Why he joined government / goal for AI leadership

• Kratsios argues that U.S. leadership in emerging technologies—especially AI—is a defining issue.
• The purpose of OSTP policy is to create a regulatory environment that:
  • allows innovation, and
  • provides guidance and oversight.
• He describes government’s potential upside and downside: it can help “steer things in the right direction,” but can also “ruin everything” if mishandled.

Background influencing his policy perspective

• He previously worked in the private sector (Scale AI) and earlier led federal technology/AI initiatives during Trump’s first term.
• He says he learned in venture/corporate environments that regulation is often the key bottleneck.
• He adds that political/state/local regulatory decisions can determine whether innovation scales.

How technology policy is actually made (decentralized, cross-agency)

• Kratsios argues policy isn’t set by a single “AI czar” or a small White House group.
• Because there is no “Ministry of Technology,” AI and tech issues are distributed across many agencies (including national security, FAA/aviation, commerce, and others).
• He uses drones as an example:
  • Commercial drone rules require FAA aviation coordination.
  • They also intersect with security concerns related to nuclear facilities.

Open-source models: addressing recent controversy

• He references a week of noise in AI policy, including letters organized by/with Y Combinator aimed at preventing restrictions on open-source models.
• He says he spoke with Secretary “Letnick.”
• He states the White House supports open source and open weights.
• He connects this to the administration’s AI strategy released last July, which he says commits to both open and closed ecosystems as mutually reinforcing.

The gap between Washington and startups

• In Washington, AI is treated as a broad societal issue—covering jobs, healthcare, data centers, economic growth, and productivity—often requiring tradeoffs across domains.
• In startup settings, founders focus on rapid iteration and the practical needs to build.
• He argues government must balance societal concerns with technical requirements—and should incorporate startup input to “do it right.”

Regulatory design: avoid hard thresholds and outdated rules

• He argues against strict “red lines” because AI evolves too quickly.
• He cites the EU AI Act as an example of rules written before major model capabilities existed (e.g., before ChatGPT), making them potentially mismatched to reality.
• He also critiques rigid compute thresholds (attributed to the Biden administration’s disclosure regime) as hard to revise once established.

Risk assessment: what developers may underestimate vs what’s exaggerated

• He highlights two major risk categories:

1. Cyber risk

   • He frames this as a “Mythos moment,” referring to dangerous cyber capabilities and the need for safeguards.
   • He notes a complication: models that can be used offensively can also be used defensively.

2. Biological risks

   • He says these have been overhyped for years and claims they haven’t been a demonstrated problem for about three years.

• Despite this, he supports building infrastructure for testing/evaluation as systems move beyond the frontier.

Countering big-tech dominance / ensuring smaller voices matter

• He agrees big tech influence can feel pervasive in Washington (“Truman Show” dynamic).
• He points to efforts to broaden participation beyond big tech, including feedback channels and mentions of a “Little Tech Association” associated with Y Combinator involvement.
• He argues a major policy threat is state-by-state patchwork regulation:
  • big companies can handle it with legal resources,
  • while startups often cannot.

Single national standard as a startup-friendly approach

• Core thesis: the U.S. should establish one national rulebook for AI to avoid fragmented compliance burdens.
• He credits antitrust actions begun in Trump’s first term (referencing cases involving Google, Apple, Facebook, and Amazon) as part of protecting competition and preventing monopolist “moats.”

Reducing compliance costs

• Kratsios says the current administration is committed to reducing regulations (citing OMB leadership).
• He distinguishes between technologies:
  • “Born free” (e.g., the early internet): new rules should be cautious or minimal.
  • “Born in captivity” (e.g., commercial drones): commercialization requires government approval, so barriers should be reduced or simplified.
• He gives examples such as:
  • drone operations, and
  • supersonic flight (e.g., Boom Supersonic), where rules like noise constraints can determine whether innovation reaches the market.

Daily role and White House policymaking mechanics

• OSTP coordinates agreement across departments through White House policy councils.
• He describes four broad policy domains/councils:
  • national security,
  • domestic,
  • economic, and
  • science/technology.
• OSTP handles science/technology topics such as:
  • AI,
  • quantum,
  • civil nuclear,
  • biotech, and
  • space.
• He explains escalation:
  • if agencies can’t agree, issues move upward,
  • potentially reaching the President for decision,
  • with the intent of limiting how often presidential time is required.

“Science: A New Golden Age” report (new emphasis)

• Kratsios discusses a newly released report proposing a modernized approach to the science ecosystem, using historical framing (FDR and Vannevar Bush’s post-WWII vision).
• He argues the structure of R&D funding has changed:
  • today the private sector does ~70% of R&D,
  • compared with the federal government doing most funding in the mid-20th century.
• Core claim: AI will transform scientific discovery workflows in fields like:
  • materials science,
  • chemistry, and
  • pharmacology.
• Vision for the future:
  • autonomous cloud/robotic labs,
  • fast-track grants,
  • agencies required to submit action plans and create incentives to draw private funding.
• Principle: “Government shapes the arena, not discoveries”—build infrastructure and incentives rather than micromanaging research.

Additional technology priorities and legislation perspective

Quantum information science

• Quantum information science is highlighted as a major but under-discussed priority.
• He references an executive order and DOE targets, including a significant quantum computer goal by 2028.

Congress vs executive action

• He argues executive orders can be repealed by a future administration, while laws cannot.
• He urges Congress to pass preemptive measures to prevent a chaotic patchwork of state AI rules.
• He also calls out needs around AI and intellectual property (IP).

Intellectual property (IP) / creator protections

• He emphasizes IP policy clarity focused on model outputs (not only training data).
• Example: if a model generates an output that imitates Mickey Mouse, he says that should be prohibited—framing this as protecting recognizable creative works/characters.
• He suggests workable licensing or revenue-sharing models may emerge through market mechanisms over time, so laws may not need to rush every detail immediately.
• He points to early market behavior through emerging deals (e.g., Yelp/ChatGPT-related licensing of reviews).

Presenters / contributors

• Michael Kratsios — Director, White House Office of Science and Technology Policy; senior science and technology advisor
• Luther — host/interviewer figure associated with Y Combinator (mentioned as coming from the audience)
• David Sachs — co-writer of the July AI strategy (cited by Kratsios)
• Secretary Rubio — co-writer of the July AI strategy (cited by Kratsios)
• Secretary Letnick — named as someone Kratsios spoke with regarding open-source policy (cited by Kratsios)
• Y Combinator / Y Combinator AI Startup School — event context (represented by the host/organizers)
• Russ Vought — mentioned as head of OMB (cited by Kratsios)
• Vannevar Bush — historical figure referenced
• Franklin D. Roosevelt (FDR) — historical figure referenced
• President Trump — referenced throughout