ANDREESSEN HOROWITZ (A16Z)

Institutional Archetype

THE LEAD INVESTOR ON BOTH SIDES OF THE TEST — Andreessen Horowitz is the investor that led the seed round of the crowd-vote model ranking, the Series A of the private-test-set benchmark company and the Series A of the evaluation-software company the labs use, took part in the Series A of the AI security-testing company OpenAI has since bought, and in March 2026 co-led OpenAI’s own round. It also gave $50,000,000 in two equal gifts to a super PAC that its founders and OpenAI’s president fund, and reports its own lobbying on “Artificial intelligence issues; open-source artificial intelligence.” Its public line is the Little Tech Agenda: “We believe bad government policies are now the #1 threat to Little Tech.” This profile scores the position, not the founders or the partners.

The finding is the interlock, not a gate. a16z grades no model. What it holds is a lead position in four companies that measure models, a lead position in the company whose models they measure, and a political program on the rules that would apply to both, each on a different page and none cross-referenced by the others.

Mandate & Origin

Self-description. The firm’s site: “a16z Capital Management is an investment adviser registered with the Securities and Exchange Commission.” Its January 2026 post on the raise: “Our mission is ensuring that America wins the next 100 years of technology. That starts with winning the key architectures of the future – AI and crypto.” The Little Tech Agenda (5 July 2024, by Marc Andreessen and Ben Horowitz): “Little Tech is our term for tech startups, as contrasted to Big Tech incumbents,” and “Our political efforts as a firm are entirely focused on defending Little Tech” (a16z, The Little Tech Agenda; a16z, Why did we raise $15B?).

Legal form. The Senate lobbying-disclosure record names the registrant “A16Z CAPITAL MANAGEMENT LLC (F/K/A AH CAPITAL MANAGEMENT L.L.C., D/B/A ANDREESSEN HOROWITZ).” The FEC lists “A16Z CAPITAL MANAGEMENT LLC” as the contributor of its two $25,000,000 gifts. The firm’s own text identifies the SEC-registered adviser by the name a16z Capital Management, L.L.C. (Senate LDA, Q4 2025 report; FEC, Leading the Future receipts).

Money raised. “At Andreessen Horowitz, we just raised over $15B. With these new funds including American Dynamism ($1.176B), Apps ($1.7B), Bio + Health ($700M), Infrastructure ($1.7B), Growth ($6.75B), and other venture strategies ($3B), we raised over 18% of all venture capital dollars allocated in the United States in 2025” (posted 9 January 2026). The post, by Ben Horowitz, is also a statement of policy purpose: “If we fail to push the policies of the country in the right direction, America will likely lose its position as the global leader in technology” (a16z, Why did we raise $15B?). Its counsel’s release gives the same fund figures and says the firm “previously advised a16z on the closing of $7.2 billion for five funds in April 2024” (Cooley).

Who runs it. Marc Andreessen is “a cofounder and general partner at the venture capital firm Andreessen Horowitz” per the Little Tech Agenda’s contributor note; Ben Horowitz is its co-author and co-founder. The funds above are managed by a partnership whose named leads on the pages read include David George and Raghu Raghuram (Growth), Martin Casado (who led the Braintrust round), Zane Lackey and Anjney Midha (AI-security and AI investing), and Jai Ramaswamy, who “oversees the legal, compliance, and government affairs functions at Andreessen Horowitz as Chief Legal and Policy Officer,” with Collin McCune and Matt Perault (authors of the December 2025 roadmap). See Marc Andreessen (a16z, The Little Tech Agenda; Braintrust; Promptfoo).

Funding & Backers

What it does not publish. The pages read give no limited-partner list, no per-fund size beyond the January 2026 figures, and no amount for any evaluator stake. Its Form D filings for individual vehicles are on EDGAR (a16z Perennial, Access and other funds filed in 2026); no filing read names an AI evaluator as an investment (SEC EDGAR full-text search, Form D, a16z).

The evaluator-layer stakes, on the record.

Beside the evaluators, the labs: OpenAI’s announcement of its 31 March 2026 round says “SoftBank co-led the round alongside a16z, D. E. Shaw Ventures, MGX, TPG, and accounts advised by T. Rowe Price Associates, Inc.” (OpenAI). See OpenAI.

Political money, FEC. The receipts of Leading the Future (committee C00916114, an independent-expenditure-only super PAC) above $1,000,000: A16Z CAPITAL MANAGEMENT LLC, $25,000,000 on 22 August 2025 and $25,000,000 on 11 February 2026; ANDREESSEN, MARC L., $12,500,000 on each of the two dates; HOROWITZ, BENJAMIN, $12,500,000 on each of the two dates; and BROCKMAN, GREG and BROCKMAN, ANNA, $12,500,000 each on 12 September 2025. The firm and its two founders gave $100,000,000 across the two dates (FEC, Leading the Future receipts; FEC, committee C00916114). See Greg Brockman.

Lobbying, Senate disclosure. a16z’s own registration reports expenses of $650,000 (Q1 2025), $790,000 (Q2 2025, with a later “2A” amendment reporting $840,000), $1,330,000 (Q3 2025) and $710,000 (Q4 2025), and $660,000, $880,000 and $600,000 for Q1 to Q3 2026. The Q4 2025 report’s issue lines include “Science/Technology: Artificial intelligence issues; open-source artificial intelligence,” “Aerospace: National Defense Authorization Act for Fiscal Year 2026” and “Health Issues: … Artificial Intelligence.” Outside registrants reporting a16z as client include BGR Government Affairs and Cornerstone Government Affairs ($80,000 a quarter each from Q2 2025 through Q3 2026) and McDermott+ (no amount printed) (Senate LDA, Q4 2025 report; Senate LDA, filings by a16z as registrant, 2025; Senate LDA, 2026; Senate LDA, outside registrants).

Institutional Voice & Intent

a16z’s register is the position paper, the comment letter and the founder’s manifesto. On AI regulation, in its words:

• Use, not development. “Policymakers will be more successful in protecting consumers if they follow historic principles of technology regulation. To ensure that the technology can achieve its potential and that Little Tech can compete with larger platforms, policy should focus on how AI is used, not how AI is built.” And: “Regulating AI models will harm startups” (27 January 2025) (a16z, Regulate AI use, not AI development).
• To the White House (14 March 2025). “a16z stands firmly on the side of Little Tech. It’s imperative that US policymakers create a regulatory framework that allows Little Tech to compete with larger companies with more resources and deeper pockets,” in comments to the Office of Science and Technology Policy’s request for information on an AI Action Plan (a16z, recommendations for the national AI action plan).
• A federal roadmap (17 December 2025). “In order for America to realize the full promise of artificial intelligence, we must both build great products and protect people from AI-related harms.” Nine pillars; the third, “Protect against catastrophic cyber and national security risks,” proposes: “One option is to direct a technical, standard-focused federal government office to identify, test, and benchmark national security capabilities—like the use of AI in chemical, biological, radiological, and nuclear (CBRN) attacks or the ability to evade human control. That work should involve consultation with independent experts and AI researchers.” The fourth: “Establish a national standard for model transparency” (a16z, A Roadmap for Federal AI Legislation).
• The agenda. “We believe American technology supremacy, and the critical role that Little Tech startups play in ensuring that supremacy, is a first class political issue on par with any other” (a16z, The Little Tech Agenda).

Stated intent: Win “the key architectures of the future – AI and crypto” for the United States, with policy that favors “Little Tech” and regulates AI “use” rather than “development.”

Observed intent: Consistent on the record: lead investor in the companies that measure and test AI, co-lead investor in the largest lab, $100,000,000 with its founders into the super PAC, and about $3.5 million of reported lobbying expenses in 2025 (the four quarterly figures sum to $3,480,000), aimed at the rule it states.

Gap: The pages that describe the evaluators say “independent” (Vals AI: “the independent evaluator of artificial intelligence”; Arena: “neutral, and community-driven evaluations”) and the pages that describe the investor say it leads their funding. The firm’s public position calls for a government office to “test, and benchmark” catastrophic capabilities “in consultation with independent experts,” while it holds the lead stake in an evaluator that sells benchmarks to the labs and a co-lead stake in a lab. No page ties a policy position to a portfolio company; no page says the stakes were disclosed at the point of each evaluation. The gap is structural and stated in four places that do not link.

Interest: a16z argues each position in its own pages. Documented, each beside a term in the record; not a claim about motive and not a verdict that the argument is wrong:

• Regulate AI use, not development. What adoption pays: the portfolio that builds models and tools, among them the OpenAI round it co-led and the evaluators it leads; rules that attach to deployments leave the model developers’ pipeline outside them. Best case, its own: “Governments have long regulated technology based on how it’s used, not how it’s made.” Who else gains: every developer large or small, and the open-source developers its lobbying line names.
• Test national-security capabilities through a federal technical office. What adoption pays: measurement infrastructure, a market in which it leads Vals AI and Braintrust and has stakes in Arena and Promptfoo; the roadmap does not say who would run the tests. Best case, its own: “capabilities should be managed based on evidence, not headlines.” Who else gains: the evaluators and the government office.
• A national model-transparency standard, narrowly drawn. What adoption pays: the disclosure that a standard would require is a task its portfolio’s evaluation and security companies sell. Best case, its own: “Just as nutrition labels provide basic information that give consumers the ability to make good choices about the food they eat, disclosing a set of ‘AI model facts’ can help people make good choices.” Who else gains: consumers and the developers that comply early.

Position in the Apparatus

• The evaluators it leads or holds, and what each says about independence. Arena, Vals AI, Promptfoo (now inside OpenAI) and Braintrust; Vals’ own pages say its results are cited in model cards from OpenAI, Anthropic, Google, Meta and xAI, and Arena’s about page says it raised funding from a16z; see The AI Evaluators.
• Seats and people. Zane Lackey and Martin Casado, general partners, are the named leads at Promptfoo and Braintrust; Anjney Midha is the named partner at Arena; no a16z partner is listed on any evaluator’s board on the pages read. A former a16z general partner, Sriram Krishnan, served as the White House’s senior AI policy adviser; see his profile for the dates and the sources. A former NSA official and Deputy National Security Advisor for Cyber and Emerging Technology, Anne Neuberger, writes on a16z’s site: “I spent the last several years working on exactly this problem from inside the U.S. government, first at the NSA and then as Deputy National Security Advisor for Cyber and Emerging Technology in the White House” (a16z, Introducing Securing America’s Infrastructure Foundation, 15 Oct 2026).
• The political network. The super PAC’s seven-figure donors on the FEC list are a16z, its two founders and OpenAI’s president and his wife; the firm is also a named corporate supporter of the Foundation for American Innovation; see Greg Brockman and OpenAI.
• Beside the larger checkbooks. The philanthropic funders of the same layer, Good Ventures, Open Philanthropy (Coefficient Giving) and the Survival and Flourishing Fund, fund the non-profit evaluators; a16z is the equity side of the commercial ones. The overlap is set out in Evaluator Alliances.

Actions & Leadership Choices

CONDUCT: STRUCTURAL — LEAD INVESTOR IN THE EVALUATORS AND THE LABS, FUNDER OF THE POLITICAL FIGHT OVER THE RULES THAT WOULD GOVERN BOTH. No wrongdoing is alleged. The structural point is that the same firm is the lead or a named investor in four companies that test or measure AI models, a co-lead investor in the lab whose models they test, a donor to the organization that runs the crowd-vote ranking, and the $25,000,000-twice donor to the super PAC that spends on AI politics, with a registered lobbying program that lists artificial intelligence. The firm’s best case is that each stake is announced on the investee’s own page, that its policy documents are public and name measurement and testing as part of the answer, and that venture investors fund new companies whether they are evaluators or labs.

Four rounds, three roles. Arena: seed lead, Series A participant, existing investor in the Series B. Vals AI: Series A lead. Braintrust: Series A lead. Promptfoo: Series A participant and earlier investor, since acquired by a lab a16z co-funds. The same firm appears as a donor on LMSYS’s page.

A founder-funded political vehicle. The FEC list shows the firm and both founders each giving the same amount on each of two dates; OpenAI’s president and his wife gave the same sum on a third. The list does not say what the money buys.

Alliance Map

These run at the same time. One firm holds the lead in a ranking, a benchmark company and an evaluation-software company, took part in the testing company a lab bought, co-leads the lab, gives $100,000,000 with its founders to a political vehicle on the same issue, and registers a lobbying program that lists AI. Every piece is on the investee’s, the FEC’s or the Senate’s own page, and none of them points to the others. The overlap is the finding.

Reach Assessment

Institutional: $15B+ raised in one announcement; lead or named position in Arena, Vals AI, Braintrust and Promptfoo; co-lead, with SoftBank, of OpenAI’s March 2026 round; $100,000,000 with its founders to one super PAC in two dates; a registered lobbying program of $600,000 to $1,330,000 a quarter. Memetic: A policy blog and a manifesto, newsletters, a podcast network, and the Little Tech Agenda posts under its two founders’ names.

Civilizational: For scale, the philanthropic funders of the evaluator layer, Open Philanthropy’s technical AI safety, security and fieldbuilding commitments were $351 million in 2025 per its own post, and the Survival and Flourishing Fund organized $63.13MM in its 2026 round (Coefficient Giving, Emily Oehlsen, Sep 9 2026; Survival and Flourishing Fund). a16z’s single January 2026 raise is many times either figure, and it sits on the equity side of the same layer.

Score note. Attack 8, defense 8 and hp 8 on first-hand receipts: the funding releases of Arena, Vals AI, Braintrust and Promptfoo, OpenAI’s round announcement, the FEC receipts, the Senate lobbying filings and a16z’s own posts. Peers: Open Philanthropy (Coefficient Giving) (8/8/9, hazard 80), the philanthropic funder of the non-profit evaluators: level on reach and insulation and one lower on durability, because a philanthropic endowment’s spend-down is stated while a venture firm’s funds renew; Good Ventures (7/9/8, hazard 80): one higher on reach, one lower on insulation; Palantir (9/9/9, hazard 90): one lower on each; LMArena (8/4/7, hazard 63), one of its investees: level on reach, four above on insulation, one above on entrenchment; Marc Andreessen (8/9/9, hazard 92.3), the co-founder, scored in the debate register: level on reach, one higher on insulation and durability because his own wealth and public voice are his own. Hazard 80.0 sits level with Open Philanthropy.

Score basis. Atk 8 is reach: lead of the Arena seed, Vals AI Series A and Braintrust Series A, participation in Promptfoo’s Series A, co-lead of OpenAI’s March 2026 round, $50,000,000 of its own money to Leading the Future and a lobbying program of $650,000 to $1,330,000 a quarter in 2025, with no statute or gate obliging anyone to use an evaluator it backs (Arena; Vals AI; FEC). Def 8 is insulation: limited-partner capital, no lab or agency named as paying it, funds with a “10 to 20 year life cycle,” offset by positions on both sides of the test (a16z, A Roadmap for Federal AI Legislation; a16z, Why did we raise $15B?). Hp 8 is entrenchment: a $15B raise announced January 2026, the largest share of United States venture dollars it reports, and standing funds; none is held in place by statute or contract (a16z, Why did we raise $15B?).

Sources: a16z, The Little Tech Agenda; a16z, Regulate AI use, not AI development; a16z, recommendations for the national AI action plan; a16z, A Roadmap for Federal AI Legislation; a16z, Why did we raise $15B?; a16z, Introducing Securing America’s Infrastructure Foundation; Cooley; Arena, seed; Arena, Series A; Arena, Series B; Vals AI, Series A; Braintrust, Series A; Promptfoo, Series A; Promptfoo, joining OpenAI; LMSYS, About; FAI, About; OpenAI; FEC, Leading the Future receipts; FEC, committee C00916114; Senate LDA, Q4 2025 report; Senate LDA, registrant filings 2025; Senate LDA, registrant filings 2026; Senate LDA, outside registrants; SEC EDGAR full-text search; Coefficient Giving, Emily Oehlsen (Sep 9 2026); Survival and Flourishing Fund, 2026