The CEOs of OpenAI, Anthropic, Google DeepMind, and Microsoft AI — the four companies building the most powerful AI systems on the planet — just signed a public letter asking Congress to make it harder for anyone to use their technology to build biological weapons.
Sam Altman, Dario Amodei, Demis Hassabis, and Mustafa Suleyman put their names on a letter organised by the Institute for Progress and the Foundation for American Innovation, calling for mandatory screening of synthetic DNA and RNA orders. The letter, published at screendna.org, warns that “there is a real possibility that the knowledge barriers which have historically prevented bad actors from obtaining biological weapons will meaningfully erode.”
Translation: our models are getting good enough at biology that the existing guardrails around genetic material aren’t sufficient anymore. Please do something about it.
🔍 THE BOTTOM LINE
The biggest AI companies are effectively telling Congress: we’re building something that could help people create bioweapons, and we can’t fix that problem on our own — the DNA supply chain needs mandatory regulation.
What the Letter Actually Asks For
Two specific policy measures:
1. Mandatory screening of synthetic DNA and RNA orders. All companies selling synthetic genetic material should be required to screen orders against databases of dangerous sequences — checking what’s being ordered and who’s ordering it.
2. Mandatory recordkeeping for traceability. Investigators need to be able to trace suspicious orders, including cases where individual sequences wouldn’t raise flags in isolation but could be combined to create something dangerous.
These aren’t radical proposals. Many of the largest gene synthesis companies already screen voluntarily — the International Gene Synthesis Consortium has been doing it since 2009. The problem is that voluntary compliance is incomplete, and the letter argues that gap is becoming existential as AI capabilities accelerate.
Why Now
The letter’s timing is deliberate and the justification is pointed. AI systems now outperform PhD-level virologists on highly technical laboratory questions, the signatories note. While they stop short of claiming an immediate threat — acknowledging the evidence is “genuinely mixed” — the trajectory warrants action now, before the knowledge barriers erode completely.
This isn’t hypothetical. In 2017, Canadian researchers reconstituted horsepox virus using $100,000 worth of mail-order DNA. No special lab. No insider access. Just money and a delivery address. And that was before AI models could accelerate the design process.
Last year, Microsoft researchers published a study in Science showing that AI protein design tools could generate potentially dangerous gene sequences that slipped right past companies’ screening software. The models suggested new protein sequences with similar structures to known dangerous ones — different enough to evade detection, similar enough to be harmful.
As Stanford microbiologist David Relman, one of the signatories, told WIRED: “AI tools enable a user to very quickly identify where to turn to order sequences that will not be subject to screening. If prompted appropriately, they can also tell you how to change the nature of your order, so that even those that are screening may be much less able to detect what it is you’re trying to make.”
The Signatories
Beyond the four AI CEOs, the letter carries weight from an unusually broad coalition:
- Patrick Collison, CEO of Stripe
- Paul Graham, Y Combinator founder
- Alexandr Wang, Meta Chief AI Officer
- David Baker, Nobel laureate and Director of the Institute for Protein Design
- Kevin Esvelt, MIT biosecurity researcher
- Christine Wormuth, former Army Secretary, now president of the Nuclear Threat Initiative
- Geoff Ralston, former Y Combinator president and partner at the Safe AI Fund
- James Diggans, VP of policy and biosecurity at Twist Bioscience
“This is a rare moment of agreement across stakeholders that are often at odds,” the letter states. “We hope policymakers will meet it with decisive action.”
The Legislative Landscape
Two bills are already in play:
- H.R. 3029, the Nucleic Acid Standards for Biosecurity Act, passed the House Science Committee by voice vote in April 2025 — but targets voluntary standards, not mandates.
- S. 3741, the Biosecurity Modernization and Innovation Act of 2026, would go further, requiring the Secretary of Commerce to issue mandatory regulations on nucleic acid synthesis security.
The letter calls for action “this session” and urges states to adopt requirements based on existing federal guidelines rather than allowing a “patchwork of conflicting laws” to emerge.
The Cynical Reading
Here’s the uncomfortable question: why are AI CEOs — companies that have been steadily weakening their own safety commitments — now asking the government to regulate a different industry instead of their own?
It’s not an unfair reading. OpenAI restructured as a public benefit corporation earlier this year, swapped its founding AGI mission for a corporate-friendly five-principle framework, and is lobbying for civilian oversight of AI safety while publicly supporting an executive order that puts the NSA in charge. Meanwhile, Anthropic has been testing models for “panic” and “anxiety” behaviours — a far cry from refusing to build capabilities that could risk humanity.
The DNA screening ask is genuinely important. It’s also conveniently targeted at someone else’s industry. It’s easier to call for regulation of biotech supply chains than to accept limits on your own model capabilities.
Geoff Ralston, one of the signatories, acknowledged this tension: “It should be very difficult, if not impossible, to ask a model to help you do something imminently dangerous.” If that’s the standard, the AI companies have work to do on their own end too.
The Honest Reading
The cynical take isn’t wrong, but it’s also not the whole picture. Dario Amodei has been warning about AI biothreats for over a year. The Mythos security crisis showed that frontier AI models can discover real vulnerabilities at scale. And the horsepox precedent — $100K and mail-order DNA to reconstruct an extinct virus — predates modern AI entirely. The DNA supply chain has been a known vulnerability for decades.
The letter’s strength is that it targets something actionable. Mandatory screening and recordkeeping are concrete, implementable, and have bipartisan support. They won’t solve the AI biothreat problem alone, but they’d close a known gap while the harder conversations about model capabilities continue.
As Relman put it: “Given that the screening may fail in some cases, we must then have other points of control. That’s where the AI companies are going to have to step up.”
Both things can be true: the AI CEOs have self-interested reasons to push for DNA regulation, and mandatory screening is still the right policy.
What This Means for New Zealand
New Zealand doesn’t have a domestic synthetic biology industry of any scale, but the supply chain is global. If US legislation mandates screening, NZ research institutions ordering synthetic DNA from overseas providers would benefit from those safeguards regardless. The AI Blueprint for Aotearoa refreshed earlier this year flagged biosecurity as an emerging AI risk, and MBIE’s AI compliance guidance covers some of the responsible use territory — but DNA screening specifically is not yet on the NZ regulatory agenda.
It should be.
❓ Frequently Asked Questions
Q: Can’t AI companies just refuse to answer dangerous biology questions? They can try, and most major labs have some biothreat safeguards. But Microsoft’s research showed AI can generate dangerous sequences that evade screening even without explicitly being asked for bioweapons. The problem is structural, not just about refusing bad prompts.
Q: Is anyone actually trying to make bioweapons with AI right now? No confirmed cases, and the letter acknowledges the evidence is “genuinely mixed.” The point is about trajectory — the barriers are eroding faster than the safeguards are improving.
Q: What’s the connection to Trump’s AI executive order? The DNA letter and the June 2 executive order are separate but related. The EO focuses on AI model evaluation before release; the DNA letter focuses on the biological supply chain. Both address the same underlying risk: AI capabilities are outpacing safeguards. Meanwhile, OpenAI released a policy paper the same week arguing that civilian agencies, not the NSA, should oversee frontier AI safety — another case of AI companies supporting regulation while trying to shape whose hands it lands in.
Q: Would mandatory DNA screening actually work? It would close the most obvious gap. Right now, not all providers screen, and screening tools aren’t catching AI-designed sequences effectively. Mandatory screening with standardized databases would raise the bar significantly, though as the Microsoft study showed, it wouldn’t be foolproof.
SOURCES
- WIRED: OpenAI and Anthropic Sign Letter to Prevent AI-Developed Biological Weapons
- The Deep Dive: CEOs Behind ChatGPT, Claude, and Gemini Warn Congress About AI Creating Bioweapons
- screendna.org — Full Letter
- ResultSense: AI leaders back bioweapon DNA-screening law
- Science: AI protein design tools generate dangerous sequences that evade screening
