Our Account

This is the account of what Perrin Institution, and Perrin & Fidutam before it, has changed outside its own walls: the legislation it has supported, the law that is now in force, the people its systems and programmes reached, the research it put into the public record, and the recognition that work received.

The measure is the one the institution argues from. A rule is judged by what happens to the people it affects. That is the test we use, and it counts everyone the rule reaches, including people who are not born yet. So every figure here is something that happened to someone outside the organisation, counted from the record it comes from: a bill number, a recorded vote, a loan issued, a scholarship secured, a paper published. Where a figure belongs to the Perrin & Fidutam years, which Perrin Institution inherits, the text says so.

Most harms can be paid for or repaired. A few cannot. Those are the ones we work on: frontier AI, engineered pathogens, machines that decide on the use of force, and attacks on systems a country cannot run without. Most of our work is on frontier AI. We publish the research. Where it helps, we also build a working version of the system a rule would apply to, so its cost can be measured instead of guessed. The record below is the evidence of how that has gone so far.

Perrin Institution has supported legislation in the Senate, the House and the New York State Legislature, with offices on both sides of the aisle.

Taken together they describe what the institution has argued for: public research infrastructure so that frontier AI is not studied only inside a few firms; warrants for data the government buys rather than compels; human review and appeal where automated systems decide things about people; rules for synthetic media, deepfakes and digital replicas; limits on AI in nuclear launch decisions; and transparency and incident reporting from the developers of the largest models. Each bill follows, with the offices we worked with, the problem it addresses and what it changes.

The RAISE Act is the law in the record. It passed both chambers of the New York State Legislature, was signed by Governor Hochul on 19 December 2025, and its final text was settled by a chapter amendment signed on 27 March 2026. Its obligations take effect on 1 January 2027.

The second state law in the United States to regulate frontier AI developers directly, after California's Transparency in Frontier Artificial Intelligence Act. It applies to developers who have trained a model above a compute threshold and who earn above a revenue threshold, and requires them to write, publish and follow a safety protocol, and to report critical safety incidents to the Department of Financial Services on a short clock. Signed 19 December 2025; a chapter amendment settling the final text was signed 27 March 2026; obligations take effect 1 January 2027.

In plain terms: from 2027, a developer that trains a model above 10²⁶ operations and earns more than $500 million a year must publish its safety protocols and report a critical safety incident within 72 hours, on penalty of up to $1 million for a first violation and $3 million for each after it. Those duties apply to the developers of the largest frontier models, the systems the institution considers the centre of the risk it works on.

A recorded vote is the plainest evidence that an argument persuaded people outside the room. The Fourth Amendment Is Not For Sale Act passed the House 219 to 199 on 17 April 2024, with 123 Republicans joining the majority, voting to extend the warrant requirement to records that agencies had been buying from data brokers instead of compelling. It is the furthest any of this work was carried in the House.

The American Data Privacy and Protection Act cleared the Energy and Commerce Committee 53 to 2 on 20 July 2022, the furthest a comprehensive federal privacy bill has advanced. The AI Training Expansion Act, which trains federal managers on what AI can do and what it risks for privacy, was reported 39 to 2 on 12 July 2023. In each case members of both parties voted for the text.

Financial exclusion is usually described as a lack of access to banks. More precisely, it is a lack of the thing banks require before access begins: a verifiable identity. A person without a birth certificate, national ID or utility bill in their name cannot open an account, cannot borrow, and cannot build the record that would let them borrow later. The requirement is reasonable on its own terms and excludes roughly a billion people.

The conventional answer is to issue documents, which is slow, expensive, and dependent on state capacity that is often the reason the documents are missing. The project took a different premise: in the communities where this bites hardest, identity is not actually unknown. It is well known locally and simply not written down anywhere a bank will accept.

So the system was designed to capture the knowledge that already exists rather than to manufacture a new document, and to run on the one piece of infrastructure that is close to universal: a mobile number.

Fidutam built that system: a SIM-card digital identity, verified by the community rather than by a document, with zero-interest microloans on top of it. It served 1,139 unbanked individuals and reached 12,615 people, and issued more than $400K in microloans across Nigeria, Uganda, Ghana, Kenya, Cameroon, and Nicaragua, with U.S. pilots in Chicago, the Bronx, and Allentown. For a person with a phone number and no papers, it offered an identity a lender would accept and credit to build a record on.

The same years produced two further programmes. The scholarship centre secured more than $3M in scholarships for low-income students by 2024. Maestro, an education and community platform connecting communities to learning and opportunity, reached 15,000 users.

Some of our research involves building a working version of the system a policy proposal describes, so that we can test what the proposal would cost and whether it can be met in practice.

Where one of those systems proves useful in its own right, we develop it into a product and keep it running. Two of the products below were deployed under Perrin & Fidutam; Arvo is the first developed under Perrin Institution.

Most denied health-insurance claims are never appealed, although most appeals succeed. The obstacles are practical: the letter is hard to read, the deadline is buried, the regulation that decides the case is one the patient has never heard of, and the appeal has to be written in a form the insurer will accept.

Arvo handles that work. A patient photographs the denial letter; the agent reads it, works out the real appeal deadline, asks the questions only the patient can answer, and drafts an appeal grounded in the regulation that governs the plan, delivered as a filing-ready PDF.

Insurers increasingly deny claims with automated systems, and the person on the other end appeals, if at all, by hand. Arvo is our reference build for the other side of that exchange: what a working right to human review looks like when the patient has an agent too.

The institution has supported legislation requiring human appeal of automated decisions, including the Algorithmic Accountability Act and the Transparent Automated Governance Act. Building Arvo is how we find out what such a right costs to exercise, and what an insurer would have to disclose for it to mean anything.

The research record runs to 383 publications, collected with the rest of the Perrin & Fidutam papers in a 1,549-page compendium that is free to read. It was produced across 10 labs, from AI governance and legal analysis to climate, foreign policy and financial infrastructure, by a network of more than 250 researchers.

The work reached government directly. 6 federal agencies were advised: the Environmental Protection Agency on climate policy research; the Department of Energy on energy infrastructure analysis; the Department of Education on educational technology policy; the Department of State on international relations policy; the Department of Justice on legal framework analysis; NASA on space technology governance. The institution's current briefings carry the same argument into frontier AI. One makes the case that explainability is not a property of a model. It is a relationship between a model and the person who has to live with its decision.

Perrin Giving funds think tanks and technical projects working on existential risk, including work we are not positioned to do ourselves.

We fund people close to the problem. We do not screen on stage or size: a small team that has already started is as eligible as an established institution.

Every grant is published with the reasoning behind it.

VenturEd Global is the first company in the portfolio. VenturEd connects early work experience with sustained mentorship and youth-led research, then follows the young people it serves to learn how early support shapes their education and careers.

Young people whose potential is overlooked deserve meaningful opportunities, and a place in the evidence that shapes their futures. VenturEd pairs early work experience with sustained mentorship and youth-led research: it helps young people build relationships, contribute to real projects and explore their interests, then follows their paths to learn how early support relates to later education, employment and professional development.

Research is central to the work. What VenturEd learns is meant to improve the opportunities open to the next generation.

The VenturEd Fellowship places young people in remote startup internships of six to eight weeks, in research, engineering, design, marketing and operations. It forms part of a two-year study across four cohorts and hundreds of participants aged 14 to 20, with follow-up that continues after the internship ends, as participants move between education, further work and the start of their careers.

The study looks at household income, school type and access to professional networks alongside different forms of mentorship: from larger groups sharing a role or project to small internships where each participant has several points of contact. It asks how group size, mentor availability, continuity of support and real responsibility relate to where young people go next.

Its questions concern career direction, further employment and education, confidence at work, and access to advice and introductions. Comparing experiences across backgrounds shows whom each approach serves, and how support might reach the people most often overlooked.

The Opportunities Project carries the same aim into community career support. VenturEd reports more than 9,000 connections to first-job opportunities, with partner job listings as the starting point for personal guidance.

Most of that guidance happens in person: talking through a role, preparing an enquiry, finding a next step. VenturEd's model for community career centres would bring it to Baltimore, Philadelphia and Washington, DC, with libraries, community centres and public universities at the heart of local access.

Outside recognition is the slowest kind of evidence and the hardest to manufacture. The work has been commended by a United States senator in an official letter, covered by the BBC, and recognised by MIT Solve, Prudential with Ashoka, the Association for Computing Machinery and the Nasdaq Entrepreneurial Center. Fidutam's financial-access work alone won more than $250K in awards.

“your initiative sets a powerful example of how passion and purpose can drive meaningful change”

Senator Chris Van Hollen, Official Letter of Commendation

Perrin & Fidutam counted more than 1,600 members in its community over the years, and Perrin Institution inherits that record. Most of the work was done by members and fellows rather than staff; the institution has only ever employed a small staff, fewer than 25 people. It has worked with institutional partners including UNESCO, UNICEF, IBM, Meta, OpenAI and Google, and fully funded its researchers' places at the Oxford Policy Conference.

Every bill number and sponsor was checked against congress.gov and nysenate.gov in August 2026. An endorsement is a vote of confidence in a proposal; it does not mean agreement with every provision or with any office's wider aims.

Figures from before 2026 are the Perrin & Fidutam record, which Perrin Institution inherits. VenturEd's figures are as VenturEd published them. Nothing on this page is a projection. To check any figure, write to contact@perrininstitution.org.