Anthropic Ties to METR Prompt Call for Congressional Investigation

Anthropic is an artificial intelligence research company founded in 2021 that develops advanced language models with a focus on safety and alignment.

The recent publication of a detailed audit by investigative blogger Kevin Bass has brought to light complex financial relationships that may influence AI policy and regulation. The audit, hosted on a GitHub repository, examines how equity from Anthropic has been directed through various nonprofit channels to organizations involved in AI safety evaluations. This has led to calls for a congressional investigation into potential regulatory capture. The claims suggest that these ties create a system where safety organizations are funded in ways that align with the interests of the AI companies they evaluate. Such arrangements could affect how risks are assessed and reported to policymakers. As AI models become more powerful, the role of independent evaluators becomes critical, yet the funding sources may compromise that independence.

The context of this audit comes at a time when AI regulation is a hot topic in Washington, with lawmakers seeking to understand the capabilities and risks of frontier models. The involvement of multiple safety organizations funded through complex channels suggests a web of influence that could affect policy recommendations. The audit points to specific documents, including the FY2025 990-PF from Good Ventures, which shows grants to Tarbell Center and other groups but no direct line to METR. This indirect approach is what makes the system resilient according to the auditor. The call for investigation is aimed at uncovering whether these ties violate any existing laws or ethical standards in nonprofit operations.

Background on the Development of Frontier AI Models

Frontier AI models represent the cutting edge of artificial intelligence capabilities, with companies like Anthropic pushing the boundaries of what large language models can do. Founded by Dario Amodei, who previously held a leadership position at OpenAI, Anthropic has positioned itself as a leader in responsible AI development. The company has released several iterations of its Claude model family, each demonstrating improved performance in reasoning, coding, and other tasks. However, the rapid advancement has also raised concerns about potential risks, including misuse for harmful purposes and the challenge of ensuring alignment with human values. These concerns have led to the emergence of various safety organizations dedicated to evaluating and mitigating these risks. The landscape is competitive, with multiple labs racing to develop more capable systems while attempting to address safety issues simultaneously.

The involvement of philanthropists like Dustin Moskovitz, co-founder of Asana and a major donor in the effective altruism community, has been significant in funding AI safety initiatives. Moskovitz has invested in Anthropic and later donated a portion of his stake to nonprofit vehicles. This donation occurred in early 2025, at a time when the valuation was lower, but with the company's growth, the value has increased substantially. The lack of public filing identifying the exact recipient vehicle has added to the mystery surrounding these funds. Such donations are part of a broader trend where wealthy individuals support both AI development and the organizations that scrutinize it. This dual role can create complex incentives in the ecosystem.

Details of the Kevin Bass Audit and Its Findings

Kevin Bass conducted an extensive review of public documents, including Form 990-PF filings from foundations, to map out the financial connections. The GitHub repository contains multiple figures that illustrate the pathways through which money and equity flow from Anthropic to METR and related groups. One key finding is that while Good Ventures Foundation does not show a direct grant to METR in its FY2025 filing, it does provide grants to organizations that partner with or support METR, such as ARC and Longview Philanthropy. The audit suggests that this indirect funding could still influence the priorities and operations of the evaluation group. Bass argues that the structure is designed in a way that makes it resistant to external oversight or shutdown, hence the term 'unturnoffable' regulatory capture machine.

The audit also examines the role of Coefficient Giving and other intermediaries in channeling funds. These vehicles allow for the distribution of resources without direct attribution to the original source. In the case of the Moskovitz donation, the under-0.8% stake in Anthropic was transferred to a nonprofit, and at the current valuation of $965 billion for the Series H round in May 2026, this could represent up to $7.7 billion. This substantial amount underscores the scale of the potential influence. The auditor emphasizes that no public filing identifies the exact recipient, which raises transparency issues. Congressional investigators may seek to obtain more details on these transactions to understand the full scope of the network.

Technical and Operational Specifics of the Alleged Ties

METR has stated that it has not accepted funding from frontier AI companies directly, but it does receive significant free tokens from companies including Anthropic, OpenAI, Google DeepMind, and Meta. These tokens are used for running evaluations on the models. Additionally, METR has partnered with these companies on various evaluation projects to assess frontier risks. The funding update from METR indicates that it raised commitments of around $71 million in the six months prior to August 2026. This funding comes from donations, and the organization maintains that it does not accept donations made by or at the direction of AI company staff. However, the audit questions whether the indirect paths through foundations linked to Anthropic equity holders could still create dependencies.

Dario Amodei has publicly proposed the use of third-party evaluators like METR for assessing the risks associated with Anthropic's models. This proposal aligns with the company's commitment to safety but also places METR in a position of influence over how risks are perceived and regulated. The partnerships for evaluations mean that METR has access to proprietary information and models, which is necessary for thorough assessments but also creates close working relationships. The audit suggests that these relationships, combined with the funding network, could lead to evaluations that are more favorable to the companies involved. The technical nature of the evaluations involves testing for capabilities that could be misused, such as in biological weapons or cyber attacks, and ensuring that models do not exhibit deceptive behaviors.

Market and Stakeholder Implications of the Allegations

The allegations could have significant implications for the AI market, particularly for how investors and stakeholders view the independence of safety evaluations. If the regulatory capture machine is as described, it may lead to calls for stricter rules on funding for evaluation organizations. Companies like Anthropic might face increased scrutiny from regulators and the public, potentially affecting their ability to secure further funding or partnerships. Stakeholders in the AI safety community may need to reassess their funding sources to maintain credibility. The $7.7 billion figure associated with the Moskovitz stake highlights the enormous financial interests at play, which could dwarf the budgets of many regulatory bodies. This disparity in resources could make it challenging for independent oversight to keep pace with industry developments.

Other stakeholders, including competitors like OpenAI and Google, may use these revelations to their advantage in the competitive landscape. The situation could also impact public trust in AI safety claims made by the companies. If evaluations are seen as influenced by funding ties, the entire field of AI governance could suffer from reduced legitimacy. Policymakers are likely to examine whether current laws on nonprofit funding and disclosure are sufficient to prevent such entanglements. The involvement of multiple foundations and giving vehicles adds layers of complexity that may require new legislation to address. Market analysts are watching how this affects Anthropic's valuation and future rounds of investment.

From a market perspective, the allegations could lead to a reevaluation of how AI companies structure their safety initiatives. Investors may demand more clarity on funding to avoid association with potential regulatory issues. The $7.7 billion stake represents a massive resource that could be used to shape the narrative around AI risks. This could result in a situation where the organizations tasked with identifying risks are supported by the very entities that might benefit from downplaying those risks. The implications extend to global AI governance, as other countries look to the US for models of regulation.

Stakeholders in the broader tech industry are also watching closely, as similar patterns could emerge in other sectors. The use of equity donations to fund related nonprofits is not unique to AI but the scale in this case is notable due to the high valuations of AI companies. Congressional investigation could lead to hearings where executives from Anthropic and METR are called to testify. This would provide a public forum for addressing the concerns raised in the audit. The outcome could influence future donations and the structure of safety funding across the field.

Expert Reactions to the Audit Findings

Reactions from experts in the AI field have been mixed, with some calling for more transparency and others defending the need for robust evaluation organizations. The quote from Kevin Bass underscores the severity of the claims, stating that the structure cannot be turned off once established. Experts in regulatory policy have noted that similar capture issues have arisen in other industries, such as finance and pharmaceuticals, where industry funding influences oversight bodies. The call for a congressional investigation is seen as a step toward greater accountability. Some safety researchers have pointed out that without adequate funding, independent evaluations would not be possible, and the current system, despite its flaws, is better than nothing. However, the scale of the Moskovitz donation raises questions about whether alternative funding models, such as government grants, should be explored.

Dario Amodei has not directly commented on the audit in the provided materials, but his previous statements on the importance of third-party evaluators remain relevant. The partnerships METR has with multiple companies suggest a collaborative approach to safety, which some view as necessary given the shared risks across the industry. Experts at organizations like ARC and Redwood Research, which are mentioned in the funding flows, may need to clarify their own independence. The Tarbell Center for AI Journalism is also listed as receiving grants, indicating that media coverage of these issues could be influenced by the same network. This aspect adds another dimension to the debate on information flow and public perception.

What Comes Next in the Investigation and Industry Response

The next steps likely involve congressional committees requesting documents and testimonies from the involved parties. The GitHub audit provides a roadmap for investigators to follow the financial trails through the various 990-PF filings and other public records. Anthropic and METR may issue statements clarifying their positions and providing additional context on the funding arrangements. METR has already stated its policy on not accepting direct funding from AI companies, which may be used to counter the allegations of capture. The industry as a whole may accelerate efforts to establish more independent funding mechanisms for safety research to avoid similar controversies in the future.

In the longer term, this could lead to the development of new standards for disclosure and conflict of interest management in the AI safety field. If the investigation proceeds, it could set precedents for how philanthropically funded AI initiatives are scrutinized. The involvement of Good Ventures and Coefficient Giving highlights the role of large-scale philanthropy in shaping the AI ecosystem. Policymakers might consider legislation that requires greater transparency for donations from AI company equity holders to safety organizations. The outcome will depend on the findings of the investigation and the responses from the key entities involved. This case illustrates the challenges of balancing innovation, safety, and regulatory integrity in a rapidly evolving technological domain.

Summary of key entities and their connections in the alleged regulatory capture network

Entity	Role in Network	Key Financial Detail
Anthropic	Frontier AI developer	Equity provided to investors like Moskovitz
Dustin Moskovitz	Philanthropist and investor	Donated under-0.8% stake valued up to $7.7B
METR	AI risk evaluator	Raised $71M commitments; receives free tokens
Good Ventures Foundation	Funding intermediary	Grants to partner orgs but no direct METR per 990-PF
ARC and Longview Philanthropy	Partner organizations	Receive grants that may flow to METR support

1. Identify the source of equity donations from Anthropic stakeholders.
2. Trace the flow through nonprofit vehicles and foundations.
3. Assess the impact on METR's independence and partnerships.
4. Evaluate the scale of the Moskovitz stake at current valuations.
5. Determine the need for legislative or oversight reforms.

I have conducted an audit of Anthropic's finances. What I have found is so shocking that I am calling for a Congressional investigation. Anthropic is not just seeking regulatory capture. It has built a regulatory capture machine that cannot be turned off.Kevin Bass, Investigative blogger/auditor