Michael Kratsios, director of the White House Office of Science and Technology Policy (OSTP), told Colossus magazine that Anthropic could fund its own AI safety work rather than ask the government to act: “Just announce tomorrow that youโre spending a billion dollars on safety”, he said, adding that the safety steps he listed have “nothing to do with the government”.
That $1 billion is about 42% of the $2.4 billion in compute that, according to the Colossus feature, Nvidia, AMD, Google, Amazon, Anthropic and OpenAI donated to the White House’s Genesis science plan, by our arithmetic.
Short explainer · Kratsios and the $1B safety line
A vertical news Short summarising this story. Figures come from the Colossus feature The Office of the Frontier; our arithmetic is labelled.
Business Pill · Least Privilege
A short explainer on least privilege, the principle of giving a system only the access it needs. It teaches the general idea only and says nothing about Kratsios or Anthropic.
The key insight: As we read the feature, the White House science office is drawing a line on who pays: Kratsios told Colossus that safety spending is a choice inside each lab, while the same office secured $2.4 billion of compute from six firms for its own science plan.
The remarks come from an interview at the White House on 15 September, reported in a Colossus feature by Harry Lambert titled “The Office of the Frontier”. Colossus says the conversation came a day after President Trump called Nvidia’s Jensen Huang at a conference to say Anthropic chief executive Dario Amodei needed to shut his mouth and that AI safety fears were a hoax.
Kratsios asked what Amodei wanted the government to do. “You control what training runs youโre doing. You control what guardrails youโre implementing”, he said, and went on to list sandbox testing, third-party evaluation firms, red teamers and whether to move engineers from the frontier to guardrails.
He also told Colossus: “The people who are โsounding the alarmโ are literally the only ones that can solve the problem”. The magazine notes that Kratsios had to quiet his own concerns after Trump’s clash with Amodei set a new mood.
AI policy moves reported by Colossus in its October 2026 feature on Michael Kratsios. The 42% ($1 billion against $2.4 billion) and the 151 days are our arithmetic.
The 30-Day Window and the Accord
Colossus reports that since 2 June the frontier labs have, under White House pressure, voluntarily shared their models with parts of the US government 30 days before release. That arrangement has run for 151 days to 31 October, by our arithmetic.
Kratsios told the magazine in August that the government was fortunate only two companies were operating at the frontier. Colossus adds that more companies have since joined it. Over the summer, it says, he drafted an executive order inspired by Demis Hassabis’s proposal for AI companies to self-regulate under the US government’s imprimatur, which Mark Zuckerberg, Jensen Huang and former AI czar David Sacks lobbied against.
At the later White House Accord on Superintelligence, the magazine reports, the executive order Kratsios was responsible for writing called for no more than renaming AI as SI. That order is the one we covered as the order telling agencies to say super intelligence. Colossus says the Accord itself was drafted by Zuckerberg and shared with Kratsios the night before.
The Business Lever: Who Pays for Safety
Kratsios’s answer puts the cost of safety on the labs’ own budgets. As he put it to Colossus, labs control whether “90% of your engineers are working on engineering the frontier, or you take 20% of them to work on guardrails”.
The same office has asked the industry for money in another form. Colossus reports the six companies pledged $2.4 billion in compute to Genesis, a platform that gives federally funded scientists access to compute and government datasets; we covered those $2.4B science compute pledges on 8 October.
It also describes the “ratepayer protection pledge”, a voluntary commitment Kratsios shepherded under which hundreds of companies agreed to build, buy or bring the power they need for data centers.
The Science Agenda
Colossus reports that staffing at the National Science Foundation is down 35% since Trump’s return, while total NIH grant spending has not fallen compared with the Biden years. It cites Kratsios’s push for metascience units with “the authority to run experiments and compel program officers to participate”.
On young researchers, the magazine says 18% of NIH’s main research grants went to lead scientists aged 36 and under in 1983, against 3% by 2010, and that the median age of first-time NIH grant recipients is now 42. “We have abandoned young scientists in America!” Kratsios said.
In June and July, it adds, four privately run nuclear reactors reached criticality under a Department of Energy pilot, above a target of three set in a May 2025 executive order Kratsios drafted.
The Structural Read
As we read Colossus, the voluntary 30-day pre-release sharing that began on 2 June is the main safety mechanism the article describes, and it rests on White House pressure rather than an order.
As we read the feature, Kratsios’s role is to write the policy after decisions are made: Sihao Huang, now at Anthropic, told Colossus that Kratsios is the staffer who stays in the room and actually writes the policy.
By our arithmetic, the $1 billion Kratsios suggested Anthropic announce is about 42% of the $2.4 billion in Genesis compute pledges Colossus reports, and the 30-day sharing has run for 151 days to 31 October.
Michael Kratsios to Colossus
“You control what training runs youโre doing. You control what guardrails youโre implementing”
Three Implications
FOR FRONTIER LABS Kratsios told Colossus that training runs, guardrails, sandbox testing, outside evaluators and red teamers are decisions each lab controls.
FOR AI POLICY WATCHERS Colossus reports that the executive order tied to the Accord on Superintelligence called for no more than renaming AI as SI, while models have been shared with the government 30 days before release since 2 June.
FOR FEDERALLY FUNDED SCIENTISTS Colossus describes Genesis as a platform giving federally funded scientists access to compute and government datasets, backed by $2.4 billion of pledged compute.
The framework says “The governance layer sits at the intersection of capability, regulation, national security, access control, and geopolitical alignment.”
As we read it, Kratsios’s remarks to Colossus place that governance layer inside the labs rather than in Washington. The framework is the owner’s argument and does not discuss Kratsios or the Colossus feature.
The quotes come from Colossus’s interviews and have not been repeated by Kratsios in an official statement that we found. Anthropic’s response is not in the article, and we did not contact Anthropic or the White House. The article does not name the government bodies receiving models 30 days before release or say which labs take part.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Michael Kratsios told Colossus that Anthropic could announce $1 billion of safety spending itself, about 42% of the $2.4 billion in compute the same six firms gave the White House’s Genesis plan, by our arithmetic, while labs have shared models with the government 30 days before release since 2 June.
A note on sourcing. This piece is based on the Colossus feature The Office of the Frontier by Harry Lambert, read on 31 October 2026. Quotes are as reported by Colossus. The percentage and the day count are our arithmetic. We did not contact Michael Kratsios, the White House or Anthropic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
