Freehold Advisory, LLC

May 19, 2026

Mr. Jack Clark, Co-founder, Head of Public Benefit, Director, The Anthropic Institute, Anthropic, PBC
CC: Mr. Dario Amodei, CEO
BCC: press@anthropic.com

By email, with acknowledgement that this letter is open.

Dear Mr. Clark:

I am the principal of Freehold Advisory LLC, an independent business serving the Pacific Northwest ecological restoration sector. Freehold’s early published research includes the whitepaper “Portfolio Complexity in Ecological Restoration” and the sector essay “Five Pressures, One Desk” (freeholdadvisoryllc.com/research). As part of the business, I am currently a Claude Pro subscriber and an active user of the Anthropic API; Claude Code is used in Freehold’s early production analytical pipeline.

I am writing to you because The Anthropic Institute’s public mandate, as launched on March 11, is “to confront the most significant challenges that powerful AI will pose to our societies.” The matter below is one such challenge, originating from a decision Anthropic itself has made.

On May 6, 2026, Anthropic announced a four-year, approximately $5 billion per year lease of compute capacity from xAI’s Colossus 1 data center in South Memphis, Tennessee — providing access to roughly 220,000 GPUs and 300 megawatts of power. The decision sits in the following documented context.

1. The Colossus facility is powered by more than 30 gas turbines that have been operating without the major-source permits required under the Clean Air Act. xAI obtained operational continuity by classifying the turbines as “temporary,” a classification that evaded the permitting process rather than completing it. The Southern Environmental Law Center filed suit; the NAACP filed a formal civil-rights complaint; the litigation is active. On May 12, 2026, xAI announced an additional 19 turbines despite the ongoing case.
2. The facility is sited in a historically Black community in South Memphis. EPA’s EJScreen tool classifies the area as overburdened on multiple pollution and demographic indicators. The area is among the highest in the United States for pediatric asthma prevalence, and reporting has linked the turbines to documented increases in local hospital admissions for air-quality-related conditions.
3. Documented annual pollution from the facility includes approximately 1,700 tons of nitrogen oxides, 500 tons of carbon monoxide, 180 tons of fine particulate matter, and 19 tons of formaldehyde — making the facility the largest industrial NOx source in the Memphis metropolitan area.
4. Reporting in WIRED and other outlets places Colossus’s annual greenhouse-gas emissions above 6 million tons CO2e and identifies Colossus as one of a small number of comparable new gas-fired AI compute campuses whose combined emissions exceed 129 million tons CO2e annually.

These facts are public, sourced, and not in serious dispute. I am happy to provide source URLs on request.

Anthropic publicly positions itself as an AI safety company whose mission is “to ensure humanity safely navigates the transition through transformative AI.” The CEO’s October 2024 essay “Machines of Loving Grace” frames AI explicitly as a tool for human flourishing, including ecological and biological health. Anthropic’s Responsible Scaling Policy commits the company to making explicit tradeoffs about deployment when third-party harms are at stake. The Anthropic Institute’s launch announcement names “the most significant challenges that powerful AI will pose to our societies” as its operating mandate. Anthropic has, on multiple occasions, publicly criticized competitor laboratories for safety-relevant operational decisions.

The Colossus lease is in tension with each of these positions.

1. “Safety company” cannot coherently mean only model-output safety. A company whose product is built on infrastructure causing acute, measurable respiratory harm to a frontline community is, in that respect, not protecting humans from harm. The choice to source compute from this facility is a tradeoff Anthropic has now made at the scale of approximately $20 billion over four years, whether or not it has been internally named as such.
2. “Machines of Loving Grace” is harder to defend when Anthropic’s compute infrastructure is producing the scale of emissions and frontline-community pollution documented at Colossus. The asymmetry between the essay’s vision of AI-enabled flourishing and the operational reality of the compute layer is a contradiction that costs reputation faster than it costs revenue.
3. The Responsible Scaling Policy governs deployment, and a lease of this scale, with this counterparty, with this documented third-party harm pattern, is a deployment decision in the policy sense. The published RSP does not currently address compute-procurement decisions explicitly. That is a gap; the gap is not exculpatory.
4. The Anthropic Institute exists, by its own founding description, to confront challenges of exactly this shape — the societal cost of AI scaling decisions. If the Institute does not have a position on this one, the Institute’s launch was a communications artifact rather than an operational commitment. You are the natural addressee for this letter for that reason.

I chose to use Claude for my business because I found it to be among the least harmful options. The Colossus lease makes that case substantially more difficult.

I am asking for these four specific, operationally tractable, publicly verifiable actions. Anthropic has previously made comparable operational commitments. For example, the public commitment to cover 100% of grid upgrade costs at new data centers. This type of ask is not foreign to the company’s recent practice.

1. Public disclosure, quarterly, of the share of Anthropic’s inference compute routed through the Colossus facility. Anthropic does not currently publish carbon emissions or compute-source breakdowns; this establishes a baseline.
   • Without this, customers cannot make informed decisions about the carbon and pollution intensity of their use of Anthropic services.
2. A public statement of Anthropic’s position on xAI’s Clean Air Act compliance at Colossus, including:
   • whether Anthropic conducted environmental due diligence prior to signing the lease, and
   • the lease’s contractual provisions regarding xAI’s regulatory compliance, and
   • the “temporary” turbine classification.
3. A public commitment that Anthropic will exit the Colossus lease in the event that xAI does not obtain valid Clean Air Act permits within a reasonable period of time.
   • Anthropic should set that period; 12 months is reasonable.
4. Publication of an adaptive “compute-procurement environmental standard” that Anthropic will apply to all future leases and partnerships, including at least: (a) disclosure of carbon intensity, (b) regulatory-compliance verification, and (c) environmental-justice site analysis. This is the standard whose absence produced the Colossus decision; publishing it forward-looking would acknowledge the gap without re-litigating the lease itself.

These asks do not require Anthropic to exit the Colossus lease today. They also do not require disclosure of proprietary compute economics. They are guardrails for Anthropic to behave, on the procurement side of its operations, as the safety-first company it has publicly committed to be.

If Anthropic has not publicly committed to all four of these asks within 90 days (Monday, August 17th, 2026), I will close my account.

I am not the customer Anthropic loses on a marginal climate cost-benefit analysis. I am the customer Anthropic loses on values consistency, a customer whose own professional credibility is partly built on structural transparency about ecological tradeoffs. There are more customers like me in Anthropic’s subscriber base than the company’s revenue concentration may make immediately visible, and we are in conversation with each other.

I would prefer to stay. I am asking Anthropic to give me reasons to stay.

Sincerely,
Ben Parker
Principal, Freehold Advisory LLC