asad ramzanali

asad ramzanali

@asad09.bsky.social

AI & tech policy, Vanderbilt Policy Accelerator. My own views.

108 followers56 following64 posts

View full thread

And the AI industry receives sales tax exemptions, property tax abatements, bonus depreciation & preferential tax treatment for investors. If AI takes jobs, the tax base shrinks further: ~3/4 of fed. tax revenue comes from taxes on labor. Now is the time to tax AI.

Today’s AI boom is actually partly funded by the tax code. OpenAI/Anthropic aren’t profitable so pay no fed. corp income tax. Alphabet, Amazon, Meta, Microsoft & Nvidia profit from AI but were 5 of the 6 largest recipients of fed. tax breaks.

News: I’m leading the Vanderbilt Policy Accelerator’s new AI Governance Project! AI is reshaping the econ, work, security & culture, but largely at the direction of private actors. AGP will build policy that centers the public's interest vanderbiltpolicyaccelerator.substack.com/p/announcing...

vanderbiltpolicyaccelerator.substack.comAnnouncing the AI Governance ProjectThe effects of artificial intelligence will be society-wide – reshaping the economy, labor, national security, culture, and even knowledge acquisition.
View full thread

We’ll hit the ground running. On deck this month: a brief on how to apply utility-style regulation to AI & a paper mapping ways to tax AI at the local, state & fed. levels. Subscribe to stay updated.

News: I’m leading the Vanderbilt Policy Accelerator’s new AI Governance Project! AI is reshaping the econ, work, security & culture, but largely at the direction of private actors. AGP will build policy that centers the public's interest vanderbiltpolicyaccelerator.substack.com/p/announcing...

vanderbiltpolicyaccelerator.substack.comAnnouncing the AI Governance ProjectThe effects of artificial intelligence will be society-wide – reshaping the economy, labor, national security, culture, and even knowledge acquisition.

We’ll focus on regulation, security & safety; AI for the public good; economic resilience; constitutional democracy; and geopolitics - building on VPA's track record of AI policy leadership. bsky.app/profile/gane...

Ganesh Sitaraman@ganeshsitaraman.bsky.social

I’m excited to announce the Vanderbilt Policy Accelerator is launching the AI Governance Project, led by VPA’s @asad09! For a few years, VPA has been working on AI policy, but at this critical moment, we’re now expanding our efforts. 1/3

Fed gov't employees are using AI to revoke visas, patient care for vets, & military ops. Last month the Nat'l Archives said most of that use of AI doesn't count as “records” under the Fed Records Act. Margaret Mullins & I analyze the guidance vanderbiltpolicyaccelerator.substack.com/p/federal-em...

vanderbiltpolicyaccelerator.substack.comFederal Employees’ AI Prompts Are Records … in Narrow CircumstancesNew National Archives guidance leaves most government AI use off the record

In short, NARA treats Chat/Claude prompts & outputs as “not received” until an employee copies them into an agency system, something few will do, effectively neutering the guidance. That's not the best reading of the statute and should be fixed

asad ramzanali reposted

The National Archives and Records Administration's new AI guidance "allows much of how the government uses AI to remain beyond the inspection of affected parties, researchers, other policymakers, and members of the public," write @asad09.bsky.social and Margaret Mullins.

justsecurity.orgFederal Employees’ AI Prompts Are Records … in Narrow CircumstancesIf federal agencies fail to create records related to their AI interactions, they effectively insulate their decisionmaking from oversight.

A couple of companies keep saying they're developing products that might be dangerous but want gov't to pace them. They can actually slow down on their own, no law needed. Ofc Congress needs to act on many fronts too. I joined @WalkerinDC on BBC to discuss the latest AI saga.

Cash, compute & data Google got from its Search monopoly became essential inputs to Gemini. And genAI is no check. It’s already concentrated & Google's vertically integrated across the whole stack: Search, Chrome & Android for distro; Gemini models, Cloud & TPUs for compute

Even in an OpenAI v Anthropic AI horserace, Google wins as investor in 1, compute supplier to both. 4 months after a judge let it keep paying Apple $20B/yr for default search, they announce Gemini will power Siri: same tactic, now in the mkt the court said would discipline Search

asad ramzanali reposted

Tech policy is only as strong as government’s ability to make it work. In our latest in Building the Future: Meet the Roosevelt Society, @asad09.bsky.social reflects on his path from Silicon Valley to the White House and why he works at “the scale of affecting the public.”

rooseveltinstitute.orgThe Race to Regulate New Tech: Q&A with Asad RamzanaliAsad Ramzanali discusses AI and tech policy, regulatory reform, and how government can deliver bold solutions in this Building the Future: Meet the Roosevelt Society interview.

asad ramzanali reposted

This wk, @joellthayer.bsky.social & I filed an amicus brief in the U.S. v. Google appeal. A federal court found Google unlawfully maintained its search monopoly, then denied remedies commensurate with the problem because genAI might discipline that monopoly (what can't AI do?!). 1/4

View full thread

And genAI isn't a competitive check. It's already concentrated, and Google is vertically integrated across the whole stack. Google has a distribution advantage: >40% of searches return an AI Overview. 3/4

This wk, @joellthayer.bsky.social & I filed an amicus brief in the U.S. v. Google appeal. A federal court found Google unlawfully maintained its search monopoly, then denied remedies commensurate with the problem because genAI might discipline that monopoly (what can't AI do?!). 1/4

But as we argue, Google's search monopoly financed & seeded its AI position: cash, compute, & data from search funded Google Brain, $650M DeepMind acquisition, & researchers who wrote the transformer paper. 2/4

Anthropic, OpenAI & Google control foundation models nearly every AI app is built on. And they're building, buying & investing in apps that compete with their customers. Tom Wheeler & I argue model providers should be barred from unjust discrimination. www.brookings.edu/articles/to-...

brookings.eduTo unleash AI innovation, stop model providers from picking the winners | BrookingsAsad Ramzanali and Tom Wheeler explain why competition is the best route to the future of inventive AI applications.

this is excellent work by @nathaliemarechal.net

Tech Policy Press@techpolicypress.bsky.social

Tech policy used to be a relatively bipartisan field, writes Dr. Nathalie Maréchal. But positions that would be defensible in a different political context aren’t at this time, she says, arguing that regardless of where you sit in industry, government or civil society, the time to choose is now.

asad ramzanali reposted

Replying to @nathaliemarechal.net

The techs we collectively call AI have use cases, but policy should be about solving problems in the public interest, not identifying ways to deploy specific technologies just for the sake of doing so. Yet that’s *still* how so many of these convos are framed. It’s exhausting. And harmful.

asad ramzanali reposted

Replying to @nathaliemarechal.net

Similarly, referring to “AI” as an actor rather than a tool serves to absolve people and institutions of accountability. I’m struck by how many so-called leaders seem happy to treat widespread “AI” adoption as a fait accompli, an act of God or natural phenomenon that can’t be contested, only managed

For months, financiers, policymakers, & technologists have warned about an AI bubble as trillions in investment chase billions in revenue. But almost no one is asking what to do after it bursts. My new paper lays out how a crash might happen & bold proposals for after the crash.

View full thread

Then Congress should go big on reforms: a Glass-Steagall for AI separating algorithms from data centers. A public cloud built from stranded data center assets. A Digital WPA for jobs. A new digital regulator. Ban extractive / surveillance business models. And more.

For months, financiers, policymakers, & technologists have warned about an AI bubble as trillions in investment chase billions in revenue. But almost no one is asking what to do after it bursts. My new paper lays out how a crash might happen & bold proposals for after the crash.

View full thread

This isn't a dot-com replay. The Mag Seven are ~35% of the stock market and borrowing at record levels. Circular financing, off-balance-sheet SPVs, and private credit link these companies to each other and the whole financial system. This is more like 2008 than 2000.

What to do after the crash: First, stop the financial engineering accelerating the crash. And prosecute related fraud: Financial crises almost always involve fraud. Accountability after 2008 was virtually nonexistent.

For months, financiers, policymakers, & technologists have warned about an AI bubble as trillions in investment chase billions in revenue. But almost no one is asking what to do after it bursts. My new paper lays out how a crash might happen & bold proposals for after the crash.

Hyperscalers plan ~$700B in capex this year – a larger GDP share than the Apollo program, Manhattan proj., highway system & nearly every other capital effort in US history. And YOU are funding it, through financial engineering that risks your 401(k), life insurance, and savings.

People assume that gov’t inaction is the best policy strategy to support Little Tech. We pulled together conservative, progressives, LT companies, investors and found 7 discrete fed. & state policies to support Little Tech that could garner bipartisan and industry support.

Those conflicts aren’t hypothetical. Last year, Anthropic cut off API access to the AI coding agent Windsurf while its own Claude Code was gaining traction, showing how model providers can act as gatekeepers and pick winners and losers.

AI Neutrality borrows from net neutrality: when firms control essential infrastructure, they shouldn’t discriminate in favor of their own services. The goal isn’t to slow AI but to protect competition & innovation for end users. Substack post: vanderbiltpolicyaccelerator.substack.com/p/net-neutra...

vanderbiltpolicyaccelerator.substack.comNet Neutrality for AIOver the course of 2025, “vibe coding” went from Silicon Valley pet projects to a global phenomenon.

asad ramzanali reposted

asad ramzanali reposted

@asad09.bsky.social and I call for better AI ambitions, beyond just business productivity. That will take forward government action and investment, not destruction. A future with new medicines for intractable diseases, better K-12 education, and improved weather forecasts is at stake.

techpolicy.pressAmerica Needs Better AI Ambitions | TechPolicy.PressArati Prabhakar and Asad Ramzanali say we are caught between fantastical claims and leaders deferring to industry without the imagination to drive progress.