@anton_d_leicht Is your point here that candidate platforms become more performative? Because I think the worry is more that the industry docs are what’s performative.
English
Andrew Freedman
640 posts
Andrew Freedman
@AndrewFATHOM
I think about public policy. A lot. I was also back-up, back-up QB for the freshman B team in high school.
가입일 Mart 2012
1.2K 팔로잉1.5K 팔로워
Andrew Freedman 리트윗함
Dean calls for "a sensible middle ground" built on "independent, nonpartisan sources of expertise that could conduct audit-like inspections of the biggest AI developers."
@deanwball, co-author of the AI Action Plan, in @TheEconomist.
Translation: Independent Verification Organizations (IVOs). Right now there is a growing bipartisan coalition building the legal framework to authorize them.
Dean W. Ball@deanwball
This weekend I am in the Economist with a piece on how policymakers should respond to Mythos with a light touch rather than a completely hands-off approach.
English
Andrew Freedman 리트윗함
🚨 NEW: Anthropic's Glasswing response shows both what responsible action looks like and why voluntary self-governance isn't enough.
Mythos, a model that can autonomously find and exploit zero-day vulnerabilities across every major operating system and web browser. Engineers with no formal security training got working exploits overnight.
Anthropic chose to act, but no framework required it, and no structure guarantees the next lab will make the same call.
We need independent verification, not labs grading their own homework. Full breakdown in our latest substack.
English
Andrew Freedman 리트윗함
1/ Virginia just became the first state to sign legislation directing a formal study of Independent Verification Organizations for AI. Bipartisan votes of 84-14 and 40-0. A unanimous bipartisan Senate vote. That’s a signal.
English
Andrew Freedman 리트윗함
this is a good next step on what I think is the most promising path: use state policy to incubate a broad and deep third-party verification ecosystem, in time to deploy it at scale when the political need for actually effective federal policy breaks through.
Fathom@Fathom_org
Governor Spanberger signed SB 384 and HB 797 into law, making Virginia the first state in the nation to advance the Independent Verification Organization framework through legislation. The bill directs JCOTS to evaluate a framework for IVOs: independent, expert-led bodies that verify whether AI systems meet safety standards. It passed with overwhelming bipartisan support, 84-14 in the House and 40-0 in the Senate. This is what smart AI governance looks like. Not government vs. industry. A better way forward.
English
Andrew Freedman 리트윗함
The momentum for what I have called “private governance” and what has taken its more mature form as “independent verification organizations” continues to grow. I applaud Virginia on taking an early step toward this idea, which I believe may be the future of frontier AI policy.
Fathom@Fathom_org
Governor Spanberger signed SB 384 and HB 797 into law, making Virginia the first state in the nation to advance the Independent Verification Organization framework through legislation. The bill directs JCOTS to evaluate a framework for IVOs: independent, expert-led bodies that verify whether AI systems meet safety standards. It passed with overwhelming bipartisan support, 84-14 in the House and 40-0 in the Senate. This is what smart AI governance looks like. Not government vs. industry. A better way forward.
English
If you only read one AI poll this year, read this one! Not just because we did it, but because we made an honest attempt to ask Americans what tradeoffs they were willing to make in AI governance.
Fathom@Fathom_org
New polling from Fathom on what Americans actually want from AI governance: ➡️ A durable, bipartisan mandate for trust and accountability ➡️ Consensus fractures when American competitiveness is at stake ➡️ Every workforce policy we tested commanded majority support ➡️ The public trusts independent experts over government or industry Full findings below.
English
Andrew Freedman 리트윗함
Our new national poll is out. Americans have a clear mandate for trust and accountability in AI, and it holds up across party lines, even when the trade-offs are made explicit.
Thank you to @ashleyrgold for the exclusive first look.
Axios@axios
Exclusive: Americans want AI guardrails but resist key trade-offs, survey finds trib.al/CLV64WF
English
What Americans, our businesses, and our institutions have just started grappling with is what to do when AI agents are autonomously or semi-autonomously making billions of choices a day that affect our economy, our livelihoods, our safety, and our wellbeing. This is more than a question of who will be liable when there is harm, which is difficult on its own. It's a deeper question of how anybody can adopt AI into their lives and be confident that it won't harm themselves or others. There will have to be a framework of trust. There are smart ways to grow that framework that will do more to aid innovation than impede it. We look forward to working with the Administration and Congress on helping to grow that framework.
English
As we have said from the beginning, preemption in the absence of a federal framework only creates greater uncertainty for our deployers, higher costs for our innovators, and more harm to families and children. This document takes a step to lay out what this certainty could look like. However, it leaves major questions unanswered — particularly how to understand the responsibilities of a technology that will be making decisions for us.
English
Andrew Freedman 리트윗함
New report: "Who Watches the Watchers"
We brought 15 experts from 6 countries to UNESCO House in Paris and spent a full day stress-testing how Independent Verification Organizations would actually work in practice.
They played government regulators, an IVO, and an AI company.
What we found 👇
English
Andrew Freedman 리트윗함
New @Fathom_org report: “Who Watches the Watchers”
How do we build a governance infrastructure for AI that actually works, that keeps pace with the technology, produces meaningful safety outcomes, and scales beyond what governments can do alone?
We brought that question to a full-day workshop at the International Association for Safe and Ethical AI (@IASEAIorg) conference in Paris.
🧵
English
We were honored to join @JoinFAI’s amicus brief. The brief makes a powerful argument that the DoW does not have the unfettered ability to harm American companies.
Foundation for American Innovation 🇺🇸🚀@JoinFAI
Today @JoinFAI filed an amicus brief in the D.C. Circuit in support of Anthropic's emergency motion for a stay—joined by @FiftyYears, @Fathom_org, China Talk, @ChorzempaMartin, @alexolegimas, @pangramlabs, @IFP, @DeanWBall, @dwarkesh_sp, @rootsofprogress, and @KhanSaifM.
English
As a sample size of one. I took it, and preferred the human-written piece four out of five times. The fifth was a toss-up, mostly because neither version did much for me. My takeaway: humans still have a clear edge when it comes to writing something deeply meaningful.
Kevin Roose@kevinroose
We made a blind taste test to see whether NYT readers prefer human writing or AI writing. 86,000 people have taken it so far, and the results are fascinating. Overall, 54% of quiz-takers prefer AI. A real moment! nytimes.com/interactive/20…
English
A law student I spoke with yesterday summed up these results in four words: "The vibes are off."
She's not wrong. Almost two years ago, our polling started to surface a key tension — people are fine with AI that works *with* them, but they'll turn against AI that feels like it's happening *to* them.
I'm not sure how to read this specific number. What I do know: the vibes are off, the reasons why matter enormously, and the real disruption from AI is barely underway.
If public mood keeps trending this direction, expect a fierce populist backlash.
Nick Field@nick_field90
NBC shows AI with worse poll numbers than ICE nbcnews.com/politics/2026-…
English
@Polymarket My take: x.com/AndrewFATHOM/s…
Andrew Freedman@AndrewFATHOM
Interesting development: In Feb, Meta patents user‑simulation “when absent…or deceased,” and now buys a social layer built for agents. Not saying this is the plan, but you can see the chessboard: persistent AI “accounts” + an agent‑to‑agent network. A new frontier for online identity, consent, & patents. @MattPRD @benparr
English
BREAKING: META acquires Moltbook, a social network built for AI agents.
English
Interesting development: In Feb, Meta patents user‑simulation “when absent…or deceased,” and now buys a social layer built for agents.
Not saying this is the plan, but you can see the chessboard: persistent AI “accounts” + an agent‑to‑agent network. A new frontier for online identity, consent, & patents. @MattPRD @benparr
Axios@axios
Exclusive: Meta acquires Moltbook, the social network for AI agents trib.al/wEZLBz0
English
This brings me back to my favorite college class, Philosophy of the Mind, with the late great @danieldennett. For those who don't know the Chinese Room thought experiment-- the idea is that if there is a man in a room who receives Chinese text, consults a manual that tells him what characters to scribble down as a result of that text, and then walks across the room and sends those scribbles out, that person cannot be said to "know Chinese". It is supposed to dispel the notion that a computer that can converse with you is therefore conscious. But, as @emollick points out in this wonderful update, the actual mechanism for using such a manual requires an unbelievable amount of time and computation. I think it lends credence to the idea that somewhere in that computation is consciousness. We might not be there yet, but I think that is what we are evolving.
Ethan Mollick@emollick
English
Catch Dean Ball on Ezra Klein for a deep, philosophical take on Anthropic/DoW and what it means for our democracy.
Samuel Hammond 🦉@hamandcheese
Your must read / watch / listen for the week: nytimes.com/2026/03/06/opi…
English