Tal Feldman

@yalelawsch | National Security | Former AI Engineer

[Washington Post Opinions @PostOpinions]

"These incidents should be a wake-up call because nations such as China do have the sophisticated tools and time to penetrate AI firms in the United States," Tal Feldman and Ibrahim Dagher write. https://t.co/F2tRTFZQUz

[Tal Feldman @talxfeldman]

As I wrote this summer in @nytimes, we should view Chinese AI models as foreign agents. More from me on this soon https://t.co/4qmoMpaTrt

[Andrew Curran @AndrewCurran_]

This is the type of thing that potentially gets new laws passed.

[Pardis Sabeti @PardisSabeti]

1/7 🌾🐄  Our letter is out in @Science: AI biosecurity must include agriculture.

Recently biological AI has brought biosecurity to the center of public debate, but most attention is on human pathogens. Crops and livestock are a major blind spot. 🧵

Gift link: https://t.co/LxiPD5gKsk

@broadinstitute @sabeti_lab @JonxFeldman @talxfeldman

[Tal Feldman @talxfeldman]

It’s ridiculously easy to hack into America’s AI companies. As I write with Ibrahim in @washingtonpost, DC must secure the labs, especially from China. No new law needed:
1) cyber standards for Pentagon AI contracts
2) export controls on models from insecure labs
3) security terms on data center loans & leases
https://t.co/jQ7tKl9Dk8

[Tal Feldman @talxfeldman]

Everyone is talking about AI bio threats in humans. What they should ALSO be talking about is biothreats in crops and livestock.

In a new letter in @ScienceMagazine, @JonxFeldman, @PardisSabeti, and I argue that agriculture deserves a large place in biosecurity conversations.

Why agriculture? Well, our screening mechanisms don't focus on these pathogens. They are also easier to experiment with, since many don't pose risks to the humans handling them. And there's lots of data about them and host defenses.

As Congress weighs the Biosecurity Modernization and Innovation Act, we're calling for:
- DNA synthesis screening that covers agricultural pathogens
- AI evaluations that explicitly test for agricultural misuse
- Early-warning disease monitoring for agriculture, not just for people

Read it here https://t.co/H12FJ1NPix

[Tal Feldman @talxfeldman]

Worried antitrust law will block AI labs from cooperating on AI safety? There's a solution that doesn't require Congress. It's called Parker immunity, and it lets any state authorize and supervise that cooperation, which shields the labs from Sherman Act enforcement. My latest in @just_security with Kevin Xiao, and Ibrahim Dagher

[Just Security @just_security]

As AI companies look to Washington for an antitrust exemption that would allow them to coordinate on safety measures, an existing legal doctrine offers another path, write @talxfeldman, Kevin Xiao and Ibrahim Dagher.
https://t.co/zb5ouMY7ML

[Just Security @just_security]

As AI companies look to Washington for an antitrust exemption that would allow them to coordinate on safety measures, an existing legal doctrine offers another path, write @talxfeldman, Kevin Xiao and Ibrahim Dagher.
https://t.co/zb5ouMY7ML

[Lawfare @lawfare]

Colocation, where multiple AI companies lease racks of servers in the same facility, carries national security and digital risks. The United States has the tools to protect this increasingly critical digital infrastructure, it just needs to use them, writes Sara Shah and Tal Feldman.

[Peter Harrell @petereharrell]

Remarkable that in a 1 year period all 535 congressional offices combined spent less on AI than a single heavy-use engineer does.

Sure, a lot of work on the Hill *should* be done by humans. But hard for these guys to understand why and how to regulate AI if they don't use it.

[Tal Feldman @talxfeldman]

Everyone is writing about the paper in @ScienceMagazine today where AI designed a working virus. My brother and I saw this coming ten months ago when the paper was out as a pre-print and wrote about it in the @washingtonpost.

The way we currently deal with biothreats is by screening against a database of known pathogens. But control-F doesn't really work for systems that generate new viruses that we haven't seen before.

We need to focus on resilience. If AI compresses the timeline for creating bioweapons, the U.S. will have to compress the timeline for responding to them. That means developing the data and AI tools needed to design countermeasures as fast as new threats appear, building facilities that can validate and rapidly produce them, and updating regulations so our responses can match the speed of these emerging risks.

https://t.co/4GaPTS94pr

[Kenneth Roth @KenRoth]

Chinese AI is cheap, but it comes with China's propaganda and censorship built in and serves Beijing over the user. It should be labeled not just "Made in China" but "Made for China."  https://t.co/FgkPEVhWZY

[Jonathan Cheng @JChengWSJ]

Tal Feldman in NYT op-ed: “The U.S. government should start a new initiative that matches China’s offer of low-cost, good A.I. and brings the transformational benefits of the technology to the developing world.”
https://t.co/cSKFnZug7g

[Matt Slotnick @matt_slotnick]

it should be obvious, but worth saying anyways: alignment is relative

[Tal Feldman @talxfeldman]

Sharing my guest essay in today's @nytimes: “The Hidden Cost of China’s Free A.I.”

China is giving away good A.I. for free, and the world is taking it up, including in Silicon Valley. But these models are built under Chinese laws that require them to promote Beijing’s interests, and researchers have found bias and propaganda baked right in. Read their internal “reasoning” and you can watch them prioritize being positive about China rather than objective.

The world is outsourcing more and more of its thinking to A.I. Imagine an employee who is brilliant, tireless, and works for free—but whose first loyalty is to Beijing, not your company or government. You would call that person a foreign agent. American companies are now hiring millions of them.

The United States should require any product built on Chinese A.I. to carry a country-of-origin label: “Made for China.” More important, Washington needs to offer the world a cheap and good alternative—especially in developing countries, where China is rapidly gaining ground.

This is a battle for the world’s mind. The United States must win it.

https://t.co/Y6nsz0kgJv

[justin curl @curl_justin]

Whoa…

“When researchers asked DeepSeek to write code for users whom the Chinese government views with hostility, such as Falun Gong or Tibetans, DeepSeek wrote code with security vulnerabilities”

Check out Tal Feldman’s essay in NYTimes today