MACHINE INTELLIGENCE RESEARCH INSTITUTE

OLYMPUS opened an institutional file on the Machine Intelligence Research Institute because it is the oldest organization in the AI-safety field and the one whose conclusion moved furthest: it began in 2000 to speed up smarter-than-human AI, spent two decades on alignment research, and now asks governments to ban the work. This is not a psychometric profile, since an institution has no Dark Triad. It is a mandate, a money record, a board and a voice. The finding is the arrangement: the people who sign the institute’s policy ask also run it, and in 2025 two of them sat on the Main Track of a funder’s recommendation round that printed a seven-figure line for the institute.

Institutional Archetype

THE ALARM WITH A DRAFT TREATY — MIRI’s program is now one demand: stop the development of superintelligence, everywhere, by agreement among states. It has the text of a treaty on its book’s site, a bestseller, a policy team that briefs legislators and an endorsement of a bill that would put the ban in statute. It also has a long record of the opposite bet. For two decades it argued the problem could be solved technically, and its own 2024 update says the alignment field will probably not move fast enough. An institution that announces that its main line of work failed, and then redirects its money and staff toward persuasion, is a rare object in this apparatus. The structural question is not whether the alarm is right. It is who sits on both sides of the table when the alarm is funded.

Mandate & Origin

In its own words. MIRI’s About page: “We do research and public outreach intended to help prevent human extinction from the development of artificial superintelligence (ASI).” Its history on the same page begins: “Eliezer Yudkowsky, along with Brian and Sabine Atkins, founded MIRI as the Singularity Institute for Artificial Intelligence with the goal of accelerating progress towards smarter-than-human AI.” The name changed in 2013: “The organization changed its name from the Singularity Institute to the Machine Intelligence Research Institute.” The page’s present position: “Unfortunately, our efforts failed to prevent the current emergency,” and “Extinction from AI is a live possibility, and the only reasonable response is to stop AI development altogether, until such a time as the alignment problem has been solved” (MIRI, About).

The legal entity, as the return prints it. Machine Intelligence Research Institute Inc, EIN 58-2565917, a 501(c)(3) with tax exemption issued April 2001 per ProPublica. The FY2024 Form 990 (signed 7 November 2025) gives the year of formation as 2000, the state of legal domicile as GA, a San Francisco mailing address and the mission as “TO ENSURE THAT THE CREATION OF SMARTER-THAN-HUMAN INTELLIGENCE HAS A POSITIVE IMPACT.” It reports 28 individuals employed, three volunteers, five voting members of the governing body and three of them independent (ProPublica, MIRI; ProPublica, FY2024 Form 990). MIRI’s own pages say it is “based in Berkeley, California” (MIRI, About). Its board, as its Team page prints it: Blake Borgeson, Nate Soares, Eliezer Yudkowsky, Anna Salamon and Edwin Evans, with Yudkowsky listed as “Co-Founder” under Leadership beside Malo Bourgon, Chief Executive Officer, Soares, President, Alex Vermeer, Chief Operating Officer, and Jimmy Rintjema, Chief Financial Officer (MIRI, Team).

The strategy shifts, in dollars and in MIRI’s words. MIRI dates its own turns: alignment as the problem around 2003; community-building about 2006 to 2012; “Agent Foundations research” from 2013; an “engineering-heavy set of alignment research directions” in 2017 to 2020; then policy and communications. Soares’s December 2020 update, quoted in the 2024 one, said the 2017 research push “has, at this point, largely failed, in the sense that neither Eliezer nor I have sufficient hope in it for us to continue focusing our main efforts there.” In January 2024 Bourgon, as CEO, set three objectives, Policy, Communications and Research, the first being to “Increase the probability that the major governments of the world end up coming to some international agreement to halt progress toward smarter-than-human AI,” and said that “policy and communications will be a higher priority for MIRI than research going forward.” The same post carries a dissent from a staff member, Lisa Thiergart: “I personally would like to note my dissenting perspective on this overall choice.” It also says “We are somewhat limited in our ability to enact policy programs by our 501(c)3 status, but we work closely with others at organizations with different structures that are more free to engage in advocacy work” (MIRI, 2024 Mission and Strategy Update).

The returns show the turn in expenses. Program expenses as the Form 990 Part III lists them:

(ProPublica, TY2020 Form 990; TY2021; TY2022; TY2023; TY2024). Outreach spending rose from $193,521 in 2022 to $1,758,513 in 2024. The research line fell and changed subject: the 2024 return lists the research team’s publications as “WHAT AI EVALUATIONS FOR PREVENTING CATASTROPHIC RISKS CAN AND CANNOT DO”, “MECHANISMS TO VERIFY INTERNATIONAL AGREEMENTS ABOUT AI DEVELOPMENT”, and “DECLARE AND JUSTIFY: EXPLICIT ASSUMPTIONS IN AI EVALUATIONS ARE NECESSARY FOR EFFECTIVE REGULATION.” In 2025, per MIRI, “The Technical Governance Team announced their new research agenda focused on the creation of an ‘Off Switch’, and drafted an international agreement to prevent the premature creation of superintelligence” (MIRI, About).

Funding & Backers

The money record, every year ProPublica has digitized. ProPublica’s chart of MIRI’s returns runs from tax year 2010; the Transparency page links returns from 2000 on. Revenue, expenses and year-end figures, as printed:

(ProPublica, MIRI; ProPublica API, MIRI; FY2023 Form 990; FY2024 Form 990). The 2021 jump is the year of two gifts MIRI announced in May 2021 (below). Since 2022 revenue has run under $2 million a year against $5.3 million to $6.9 million of spending, and net assets have fallen each year. In 2024 contributions were $595,907, investment income $894,015 and other revenue $44,991. Grants and similar amounts paid were $280,536 in 2024 and $318,000 in 2023; Schedule I lists the 2024 amount as one research grant of $119,058 to one individual and “PRIZES AND AWARDS” of $161,478 to 704 recipients (ProPublica, FY2024 Schedule I).

Pay, as Part VII prints it (reportable compensation from the organization, with titles as each return prints them):

The highest-listed employee on the 2019 return is Edward Kmett at $428,513 and on the 2021 return $599,902 (ProPublica, TY2019; TY2021). The FY2024 return lists Scott Garrabrant at $296,735 with $222,834 of other compensation and Abram Demski at $169,915 with $132,111. Rintjema, Treasurer and CFO, is listed with $0 reportable compensation and $117,221 of other compensation, and Schedule F lists “COMPREHENSIVE ADMINISTRATIVE SERVICES” in North America outside the United States, “OFFERED ON AN INDEPENDENT CONTRACTOR BASIS,” at the same $117,221 (ProPublica, FY2024 Schedule F). Schedule O: “THE BOARD OF DIRECTORS MEMBERS ARE REQUIRED TO RECUSE THEMSELVES FROM VOTING ON ISSUES FOR WHICH THEY HAVE A CONFLICT OF INTEREST” (ProPublica, FY2024 Schedule O).

Donors, as MIRI states them. The Transparency page: “Since its founding in 2000, the Machine Intelligence Research Institute has had thousands of distinct donors. Our largest contributors, each of whom has given at least $500,000, are:” an anonymous Ethereum cryptocurrency investor in 2018 and 2021; Open Philanthropy, “a joint initiative with the philanthropic foundation Good Ventures”; Vitalik Buterin, “the inventor and co-founder of Ethereum”; “A different anonymous Ethereum cryptocurrency investor in 2026”; Jaan Tallinn, “Skype and Kazaa developer and co-founder of the existential risk institutes FLI and CSER”; the Thiel Foundation, “a private foundation funded by Paypal co-founder and venture capitalist Peter Thiel”; “A different anonymous Ethereum cryptocurrency investor in 2017”; the Berkeley Existential Risk Initiative; Jed McCaleb; Loren Merritt; Edwin Evans, “Cofounder of Linc Global”; and Richard Schwall (MIRI, Transparency). The list names no AI developer and no evaluator. Evans is also listed on MIRI’s board and as chairperson on the returns through 2022 and as a board member in 2023 and 2024.

The large gifts, in MIRI’s announcements.

• April 2020, Open Philanthropy (now Coefficient Giving). “MIRI has been awarded a two-year $7,703,750 grant by Open Philanthropy — our largest grant to date,” which with the second half of its 2019 grant “amounts to $4.38M per year over two years, or roughly 60% of our predicted 2020–2021 budgets”; the post credits “$1.46M” to “Open Philanthropy’s new partnership with Ben Delo, co-founder of the cryptocurrency trading platform BitMEX,” and carries an update: “Open Philanthropy decided not to move forward with its partnership with Ben Delo. This doesn’t affect the size of the grant we received, but means that other funders provided the funds for Open Phil’s grant.” The same post announces “a $300,000 grant from the Berkeley Existential Risk Initiative” and “a $100,000 grant by the Effective Altruism Funds Long-Term Future Fund” (MIRI, 27 April 2020). Coefficient Giving’s rename: “Today, Open Philanthropy is becoming Coefficient Giving” (Coefficient Giving).
• May 2021, two gifts. “This anonymous donation comes from a cryptocurrency investor who previously donated $1.01M in ETH to MIRI in 2017”; it was 3001 MKR, “MIRI liquidated the donated MKR for $15,592,829 after receiving it,” subject to a rule that MIRI “can spend a maximum of $2.5M of the gift in each of the next four calendar years, 2021–2024,” with “The remaining $5,592,829” available in 2025. Second, “the inventor and co-founder of Ethereum, Vitalik Buterin, yesterday gave us a surprise donation of 1050 ETH, worth $4,378,159” (MIRI, 13 May 2021).
• 2025 fundraiser. MIRI said “Thanks to generous support we received in 2020 and 2021, we didn’t need to run any fundraisers in the last six years.” The December 2025 fundraiser, “MIRI is running its first fundraiser in six years, targeting $6M. The first $1.6M raised will be matched 1:1 via an SFF grant,” closed with “just over $1.8M.” It states its reserve goal (“Our goal at MIRI is to have at least two years’ worth of reserves on hand”), projects about $7.1M of 2025 spending and “a median of $8M” for 2026, and says of the book: “unless the book dramatically exceeds our expectations, we won’t ever see a dime” (MIRI, 2025 Fundraiser).

The Survival and Flourishing Fund tables. The Survival and Flourishing Fund’s recommendation pages print MIRI as follows. SFF-2024 (Funder Jaan Tallinn): “Machine Intelligence Research Institute,” Freedom track $54,000, “General support,” with a speculation grant of $40,000. SFF-2025 (Funder Jaan Tallinn): “Machine Intelligence Research Institute (MIRI),” Main $46,000, Freedom $1,561,000, total $1,607,000, with “Matching: {$1,392,000}” and a speculation grant of $215,000. MIRI also appears as a receiving charity for other applicants’ grants: the Eisenstat Research Program ($593,000 in 2025, split half to MIRI and half to Ashgro; $500,000 in 2026, Funders Tallinn and Dustin Moskovitz) and AI Impacts ($351,000 in 2026, “General support of AI Impacts”); in 2024, “Eisenstat Research Directions” ($103,000) went to MIRI and Ashgro (SFF-2024; SFF-2025; SFF-2026). MIRI’s fundraiser calls the 2025 match “a $1.6M matching grant from the Survival and Flourishing Fund” (MIRI, 2025 Fundraiser).

Institutional Voice & Intent

In MIRI’s January 2024 account the aim is a halt by agreement: “Policy efforts like this seem very unlikely to save us, but all other plans we know of seem even less likely to succeed.” Its 2025 fundraiser puts the thesis in the book’s words: “If any company or group, anywhere on the planet, builds an artificial superintelligence using anything remotely like current techniques, based on anything remotely like the present understanding of AI, then everyone, everywhere on Earth, will die,” and “The default outcome is lethal, but the situation is not hopeless; machine superintelligence doesn’t exist yet, and its creation can yet be prevented.” (MIRI, 2024 update; MIRI, 2025 Fundraiser).

• The book. If Anyone Builds It, Everyone Dies, by Yudkowsky and Soares, published by Little, Brown and Company (Hachette Book Group). MIRI reports: “It was an instant New York Times bestseller, reaching #7 on the nonfiction list,” with blurbs “from the national security establishment” and “from top scientists and academics (e.g., George Church, Yoshua Bengio, and Ben Bernanke),” and endorsements “from Whoopi Goldberg to Matt Yglesias to Steve Bannon.” On the first anniversary MIRI wrote that the book “was held up in the chambers of Congress by Representative Brad Sherman in January” and “We think the book aged well (unfortunately)” (MIRI, 2025 Fundraiser; MIRI, One Year Closer, 16 September 2026). The reception, both ways, is on MIRI and the Off-Switch Problem.
• The draft treaty. On the book’s site, “A Draft of a Treaty, with Annotations”: “We are not policymakers and we are not well-versed in international law.” Article text: “Treaty Parties hereby establish the International Superintelligence Agency (ISIA)”; “Any training run exceeding 1e24 FLOP or any post-training run exceeding 1e23 FLOP” is banned; a covered chip cluster is “any set of AI chips or networked cluster with aggregate effective computing capacity greater than 16 H100-equivalents.” The authors call the numbers “our best guess” (If Anyone Builds It, Everyone Dies, draft treaty).
• The Ban Artificial Superintelligence Act of 2026. MIRI’s position, dated 23 September 2026 and “endorsed by Bourgon, Soares, and Yudkowsky on behalf of MIRI”: “We hereby endorse the Ban Artificial Superintelligence Act of 2026 because it directly confronts the extinction threat that humanity is facing and would codify the primary policy goal we think the world needs: a ban on the development of superintelligence.” It asks for more than the bill gives: “The bill lacks tracking and monitoring of AI chips.” It also says “Many MIRI staff want Earth’s civilization to build ASI eventually, after the point where ASI can be created in a fashion known in advance to be non-ruinous” (MIRI, Ban ASI Act position). The AP’s account the same day: the legislation “would permanently prohibit artificial superintelligence, which it defines as a system that exceeds human cognitive ability or has sufficient capabilities to plan and execute the destruction or disempowerment of humanity,” “Violations could carry significant penalties, including up to 20 years in prison in some cases,” and “The bill from Sanders and Casar faces long odds in the Republican-controlled Congress” (Associated Press via Vermont Public). MIRI’s 30 September newsletter calls it “shockingly good” and adds: “MIRI’s Technical Governance Team has been briefing members of Congress in the wake of the rogue AI events, consulting on legislation, recruiting policy advisors, and running a research fellowship program” (MIRI, Newsletter #127).

Stated intent: Prevent human extinction from artificial superintelligence by getting governments to stop its development until the alignment problem is solved.

Observed intent: Move the institute’s money, staff and public presence from alignment research to persuasion and governance design. Outreach spending went from $193,521 (2022) to $1,758,513 (2024), the research line now lists governance papers, and the institute has put a treaty text, a bestseller and a bill endorsement on the record in the year since the book.

Gap: The record shows no gap between the stated aim and the activity. The gap that exists is inside the institute’s own papers. Its 2024 research paper on evaluations says of them “nor should they be relied upon to definitely trigger before a catastrophic incident could occur” and “we can’t rely on them as our main way of ensuring AI systems are safe”; the Act it endorses has training pause “if a model develops precursor characteristics, such as the capacity to automate AI R&D or the capacity to avoid human shutdown attempts,” a trigger that has to be measured. MIRI states the first view and endorses the second in documents ten months apart, and its endorsement says “The Act is excellent but not perfect.” The record does not settle whether the two sit together; for the arrangement it does not need to (Barnett and Thiergart, MIRI; MIRI, Ban ASI Act position).

Interest: MIRI argues for policy positions in its own words; what adoption would pay it is set out from its returns and pages. No fee, contract or procurement mechanism is documented for any of them.

• A statutory ban on developing superintelligence. Position: the endorsement quoted above. What it pays: no payment is documented. The documented product is demand for the institute’s output: its fundraiser says its governance team must “extend that earlier work into concrete, tractable, shovel-ready packages that can be handed directly to concerned politicians and leaders,” and a statute with a federal agency and an international policy creates the readers for that work; the same fundraiser projects “$3.2M Outreach” and “$2.3M Research” for 2026. Against any reading of payment, MIRI states that “book sales will not be a source of net income for us.” Best case, MIRI’s: “We do not mean that as hyperbole. We are not exaggerating for effect.” Convergence: the sponsors, Sanders and Casar, and employees of AI developers named in the AP’s account as supporting the bill; and the evaluator layer, since the Act’s pause trigger and pre-deployment approval need capability measurement of the kind METR, Apollo Research and the labs’ other testers run (MIRI, 2025 Fundraiser; Associated Press via Vermont Public).
• An international agreement with compute caps and chip monitoring. Position: the draft treaty and the endorsement’s line that “In line with the Act, we also strongly believe it should be ’the policy of the United States to prevent the development of artificial superintelligence globally.’” What it pays: the treaty creates an agency and a verification regime of the kind MIRI’s Technical Governance Team says it studies (“Mechanisms to verify international agreements about AI development” is on its 2024 return); the fundraiser reports the team “regularly provides input on RFCs and RFIs from various governmental bodies” and took part in “the EU AI Act Code of Practice Working Groups.” Best case, MIRI’s: “Many of the quantities and numerical thresholds we use in our draft constitute our best guess, but they should still be treated only as guesses.” Convergence: MIRI’s own newsletter lists the factions that the same ask now reaches: “Senators Sanders and Warren have called for a halt to frontier AI development,” “Senator Hawley launched an investigation into ’the existential risk of new AI products,’” “Steve Bannon spoke at the Pro-Human Assembly,” and Pope Leo XIV “has warned” about control; the converging interests are the finding (If Anyone Builds It, Everyone Dies, draft treaty; MIRI, Newsletter #127; MIRI, 2025 Fundraiser).

Interest is a documented fact about each position, not a claim about motive, and not a verdict that the argument is wrong. Each best case is stated at the strength its author states it.

• Social accounts and feeds. X @MIRIBerkeley; RSS intelligence.org/feed.

Position in the Apparatus

MIRI sits at three junctions of the evaluator layer’s map: it is a recipient of the same funders, its officers have sat where funding is recommended, and its own research is a critique of the layer’s main product.

The same funders. One funder, Jaan Tallinn (with Dustin Moskovitz from 2026), is the printed Funder on the Survival and Flourishing Fund lines for MIRI and for the evaluators. Totals as the tables print them:

(SFF-2025; SFF-2026). The funder named in MIRI’s own donor list, Open Philanthropy, is the funder of record for the same layer: see Open Philanthropy and The EA to AI-Safety Funding Pipeline, which carry its grant lines for Apollo Research, Redwood Research, GovAI and Epoch. Good Ventures, named on MIRI’s Transparency page as the foundation behind Open Philanthropy, is the same source.

The committee. The SFF-2025 announcement states: “In 2025, twelve people across three tracks participated as ‘Recommenders’ (grant evaluators) in a single round of a grant-recommendation process for Funder Jaan Tallinn,” and lists, among the public names, “Malo Bourgon (Main Track)” and “Nate Soares (Main Track),” MIRI’s Chief Executive Officer and President, with the Main Track’s other public names Nathan Labenz and Seán Ó hÉigeartaigh. The same page prints MIRI’s line of $1,607,000. It describes the process in the Fund’s words: “Funders retain the right and freedom to make and/or recommend grants that the S-Process, as a whole, did not endorse,” and “the S-Process structure doesn’t impose restrictions on Recommenders and Funders communicating with each other as much as they want outside of the S-Process.” The 2026 round’s public Recommenders include Benjamin Weinstein-Raun, whom Palisade Research’s page describes as acting director of AI Impacts who “has been a member of the technical staff at SecureDNA, Redwood Research, MIRI, and Cruise Automation”; the 2026 table prints AI Impacts at $351,000 with MIRI as legal recipient (SFF-2025; SFF-2026; Palisade Research, About).

The board. Blake Borgeson is a MIRI director and a board member of Palisade Research; Palisade’s page says “Blake helped lead the Machine Intelligence Research Institute for four years and remains on both MIRI’s and Recursion’s boards” (Palisade Research, About; MIRI, Team). Palisade’s Eli Tyre “supports grantmaking at Jaan Tallinn’s Survival and Flourishing Fund” and did “contract work for Lightcone Infrastructure, the Machine Intelligence Research Institute, and the Berkeley Existential Risk Initiative” per the same page.

Its own account of the evaluators. MIRI’s research team published its view in November 2024: “Evaluations can establish lower bounds on AI capabilities and assess certain misuse risks given sufficient effort from evaluators,” but “evaluations face fundamental limitations that cannot be overcome within the current paradigm,” and “we should not rely on them as our main way of ensuring AI systems are safe.” Its recommendations include “Third-party audits with sufficient affordances”: “Auditors should be given the same access to AI systems as the AI developers (e.g. fine-tuning, base models, helpfulness-only models, state-of-the-art AI agent scaffolds),” and “Conservative red lines,” “commitments to halt AI development if a capability evaluation is triggered.” The institute that is funded by the evaluators’ funders thus argues for more outside testing with fuller access and for treating it as a floor, not as the safety case (Barnett and Thiergart, MIRI). See Who funds the evaluators.

Actions & Leadership Choices

• 2023. The strategy turn begins; MIRI’s account lists “my participation in one of the U.S. Senate’s bipartisan AI Insight Forums”; its return calls it “THE U.S. SENATE’S EIGHTH BIPARTISAN AI INSIGHT FORUM” and lists “YUDKOWSKY’S TIME OP-ED.” Bourgon becomes CEO in October (MIRI, 2024 update; ProPublica, TY2023 Form 990).
• January 2024. The strategy update, above. It names communications and policy as “a higher priority for MIRI than research going forward.”
• 2024 and 2025. The Technical Governance Team, per MIRI, “participated in the EU AI Act Code of Practice Working Groups” and “regularly provides input on RFCs and RFIs from various governmental bodies” (MIRI, 2025 Fundraiser). See EU AI Act.
• September 2025. The book, “If Anyone Builds It, Everyone Dies,” with the draft treaty “to coincide with the book release” (MIRI, 2025 Fundraiser).
• Late November 2025. “Malo was invited to provide testimony to a committee of the Canadian House of Commons; and TGT researcher Aaron Scher was invited to speak to the Scientific Advisory Board of the Secretary-General of the UN on AI verification as part of an expert panel” (MIRI, 2025 Fundraiser).
• December 2025. First fundraiser in six years, $1,802,437 raised against a $6M target with a $1.6M SFF match.
• 2026. The Transparency page adds “A different anonymous Ethereum cryptocurrency investor in 2026” to its list of donors of at least $500,000 (MIRI, Transparency).
• 23 and 30 September 2026. The endorsement of the Ban Artificial Superintelligence Act of 2026 and the newsletter that reports the Technical Governance Team “briefing members of Congress in the wake of the rogue AI events” (MIRI, Ban ASI Act position; MIRI, Newsletter #127).

CONDUCT: STRUCTURAL — THE PEOPLE WHO WRITE THE ASK RUN THE INSTITUTE AND SIT ON THE COMMITTEE THAT RECOMMENDS ITS MONEY. The documented facts are plain and public: MIRI’s officers sign its policy ask; the same CEO and President sat on the Main Track of a round that printed $1,607,000 and a $1,392,000 matching figure for MIRI; the institute’s research budget has moved to governance as its reserves fall. Nothing in the record says the committee’s recommendation depended on the seat. The point is the dependency graph, which is short.

Alliance Map

These run at the same time. One institute is funded by the philanthropy that funds the evaluators, has its CEO and President on the committee in a round that funded it, publishes a paper on the limits of the evaluators’ product, briefs legislators, testified abroad and endorses a bill. Each tie is stated by MIRI, a filing or the funder’s page. The overlap is the finding.

Reach Assessment

Institutional: A policy shop, a bestseller and a draft treaty that legislators and the press cite, with no statute behind it. Memetic: “If anyone builds it, everyone dies” is the one sentence of the field that reached the general reader, by MIRI’s account a New York Times bestseller and the subject of a bill. Civilizational: If its premise is right, nothing else in this apparatus matters; if it is wrong, the institute is the origin of a regulatory demand that a Department of Artificial Intelligence and a pause trigger would implement. The structural exposure is not the argument. It is that the argument’s funders, authors, officers and recommenders are a short list, and the list is printed.

Score basis. Atk 6 is reach: MIRI’s fundraiser says “the bulk of our engagement with policymakers and national security professionals has been done either by our CEO (Malo Bourgon), our President (Nate Soares), or the TGT researchers themselves,” the September 2026 endorsement is “endorsed by Bourgon, Soares, and Yudkowsky on behalf of MIRI,” the book was “reaching #7 on the nonfiction list,” and the SFF-2025 page lists Bourgon and Soares as Main Track Recommenders in the round that printed MIRI’s $1,607,000 (MIRI, 2025 Fundraiser; MIRI, Ban ASI Act position; SFF-2025). The points under 7 are that no statute, procurement rule or release gate obliges anyone to act on it, that the bill it endorses faces what the AP calls “long odds,” and that the figures it names for its own reach are its own. Def 7 is funding insulation: net assets of $15,242,215 at the end of 2024 and a stated goal of “at least two years’ worth of reserves,” donors named on its own page and none an AI developer; held under 8 because revenue was $1,534,913 against $6,508,701 of spending and part of the recent money is a fund whose committee included its CEO and President (ProPublica, FY2024 Form 990; MIRI, Transparency). Hp 7 is entrenchment: exempt since 2001 and filing every year since, with a treaty text and a book that persist; against that, nothing in statute or procurement installs it and net assets are about half their end-2021 level (ProPublica, MIRI). Peers on the same values: the Alignment Research Center at 6/7/7, hazard 60, holds the same numbers as a research shop with a spin-out, and this file sits four points above it because the MIRI record adds funder-committee seats; Redwood Research at 6/8/7, hazard 66, holds a point more insulation; the Survival and Flourishing Fund at 6/8/6, hazard 67, is the funder on the same tables with a point more insulation and a point less entrenchment; Palisade Research at 7/6/6, hazard 58, holds a point more reach on the evaluation side and a point less insulation. The values hold.

Sources: MIRI, About; MIRI, Team; MIRI, Transparency; MIRI 2024 Mission and Strategy Update; MIRI’s 2025 Fundraiser; MIRI’s Largest Grant to Date (27 April 2020); Our all-time largest donation, and major crypto support from Vitalik Buterin (13 May 2021); MIRI’s Position on the Ban Artificial Superintelligence Act of 2026; MIRI Newsletter #127; If Anyone Builds It, Everyone Dies: One Year Closer; A Draft of a Treaty, with Annotations; If Anyone Builds It, Everyone Dies: Hachette Book Group; What AI evaluations for preventing catastrophic risks can and cannot do: arXiv; ProPublica Nonprofit Explorer: Machine Intelligence Research Institute Inc; Form 990, tax year 2024: ProPublica; Form 990, tax year 2023: ProPublica; SFF-2024 Recommendations; SFF-2025 Recommendations; SFF-2026 Recommendations; Open Philanthropy Is Now Coefficient Giving; Palisade Research, About; Sen. Bernie Sanders unveils bill to ban artificial superintelligence: AP via Vermont Public.