On 4 June 2026, the chief executives of OpenAI, Anthropic, Google DeepMind and Microsoft AI signed an open letter asking Congress to make screening of synthetic DNA orders mandatory. They were four names on a list of more than 120, alongside two of the 2024 Nobel laureates in chemistry, the heads of the American companies that actually manufacture synthetic DNA, and former senior biodefence officials. The letter's claim is narrower than the headlines it produced. It says AI systems now "outperform PhD-level virologists on questions about highly technical laboratory procedures", and that there is "a real possibility" the knowledge barriers that have historically kept biological weapons out of reach "will meaningfully erode". It also says, in the same breath, that the evidence about what this means for present-day threats is "genuinely mixed".
When Sam Altman, Dario Amodei, Demis Hassabis, and Mustafa Suleyman sign the same document, it carries unusual weight. These four are the people responsible for building the systems they are warning about.
The Ask: Screen Every DNA Order
The letter, co-organised by the Foundation for American Innovation and the Institute for Progress, makes a narrow ask. Providers of synthesised DNA (companies that will synthesise custom sequences and mail them to you), and the manufacturers of the benchtop machines that make it, should be required to check synthesis requests for sequences of concern and to verify that a customer is legitimate before shipping. They should also keep records of orders and sequence data, so that anything the initial screen misses can be traced back to its source. None of that is required by statute today. The main industry scheme, run by the International Gene Synthesis Consortium since 2009, is voluntary, and the letter notes that many of the largest providers already screen and keep records of their own accord.
The underlying claim is that AI has changed the calculus. Designing a dangerous pathogen historically required deep expertise in microbiology, virology, and molecular biology, the kind of knowledge that took years to accumulate and was hard to acquire outside an institution. AI systems that can reason about protein structure and genetic sequences plainly erode part of that barrier, though how much is contested. The letter puts it as AI systems that "outperform PhD-level virologists on questions about highly technical laboratory procedures", while calling the evidence on present-day threats mixed. Amodei makes the sharper version of the argument in his own name: in The Adolescence of Technology (January 2026) he names biology as the misuse risk he worries about most, and calls for exactly the mandated gene synthesis screening this letter asks for.
The letter does not claim that any current AI system can produce a bioweapons blueprint on demand. What it claims about timing is procedural: the need is urgent given the pace of change, Congress should act this session, and states should build on existing federal and industry guidelines rather than write a patchwork of their own.
Why the Executives Point Downstream
The position these four executives have put themselves in is structurally awkward. The letter's argument runs: our systems make dangerous capabilities more accessible, so please regulate the downstream supply chain to limit the damage. The argument they decline to make, and that nobody is forcing them to make, is that the development of systems with these properties should itself be governed.
The pattern is familiar. The same competitive dynamics that drive capability development also produce a particular flavour of safety advocacy, one focused on mitigating the downstream effects of dangerous capabilities rather than limiting the capabilities themselves. Export controls on chips, screening of DNA vendors, watermarking of AI-generated content: all of these target the periphery while the capabilities keep scaling.
None of this implies bad faith. Amodei in particular has been among the more serious voices on AI risk for years. But the shape of the advocacy is telling. The competitive dynamics that make unilateral restraint individually irrational also shape which safety positions executives are willing to advocate in public.
AI and Dual-Use Biology
The biosecurity concern predates this letter. The International AI Safety Report 2026, published on 3 February 2026 by more than 100 experts under Yoshua Bengio's chairmanship, records that general-purpose models can supply expert-level laboratory information, and that several developers shipped 2025 models with extra safeguards because they could not rule out that those models would help a novice. The report does not rank that risk above the others it covers, and it says plainly that how far physical barriers still constrain a would-be attacker is difficult to judge. The worry isn't new. The pace is.
AlphaFold, Google DeepMind's protein structure prediction system, was a real scientific breakthrough; the company says it has predicted more than 200 million structures and that its database has been used by over 3 million researchers. It also made it easier to reason about protein function in ways that carry dual-use implications. The same goes for large language models that have absorbed vast amounts of biological literature and can synthesise it on request. The beneficial and harmful applications run on the same underlying capability. You can't deploy "the part that helps with cancer research" while withholding "the part that could inform pathogen enhancement."
Screening the DNA vendors is a sensible downstream control that leaves this underlying dynamic untouched. It is probably worth doing anyway.
The Legislation Already in Motion
The letter isn't arriving into a policy vacuum. On 29 January 2026, Senators Tom Cotton and Amy Klobuchar introduced the Biosecurity Modernization and Innovation Act of 2026 (S.3741), which would direct the Secretary of Commerce to require gene synthesis providers to screen their orders and customers, and would set up a biotechnology governance sandbox at NIST. The joint executive letter looks designed to give that bill industry momentum, and tech-sector support for biosecurity regulation is an unusual alignment. It has not yet translated into legislative movement: as of 16 October 2026 the bill has had no action since it was read twice and referred to the Senate Commerce Committee on the day it was introduced, although its cosponsor list has grown to nine, most recently on 22 September 2026.
Whether screening DNA orders addresses the actual risk depends on what you think the risk is. If the threat is someone using AI to design a dangerous sequence and then ordering the physical DNA to synthesise it, vendor screening is a meaningful control, though not a complete one. A Microsoft-led study published in Science on 2 October 2025 reported that open-source protein design tools could rewrite a toxic protein's sequence so that most screening systems would probably miss it. The work was done in simulation, and the screening software was patched before the paper appeared. If the threat is AI giving non-state actors enough technical uplift to build biological threats by routes that never involve ordered synthesis, vendor screening is a partial solution to a portion of the problem.
Two Threats, One Bill
The signatories are not naive about what they're building. Amodei's January 2026 essay on the risks of powerful AI is unusually direct for a frontier lab chief executive, and it argues for the screening mandate this letter asks for. Hassabis co-founded Google DeepMind and has spent years talking about AI risk in public. When the people running OpenAI, Anthropic, Google DeepMind and Microsoft AI tell Congress that their own field may erode the barriers to biological weapons, and the companies that manufacture synthetic DNA sign the same page, that carries weight because of who is making the argument.
It also sharpens the stakes. For years the risk calculus around advanced AI has been dominated by the alignment problem: what happens if a highly capable system pursues misaligned goals. The biosecurity letter points somewhere else, at what happens when humans use AI capabilities to pursue destructive goals. These two risks call for different responses. Alignment has to be solved on its own terms; misuse demands governance of the AI and the infrastructure around it.
Neither is solved today. Screening synthetic DNA vendors, if a bill ever passes, addresses a narrow slice of the misuse problem. The alignment problem barely registers on any legislative agenda. The closest thing to a foothold is California's Transparency in Frontier Artificial Intelligence Act, signed in September 2025, which makes the largest developers publish how they test for catastrophic risk, including a model circumventing their own guardrails, and report serious incidents to the state. That is disclosure, not a solution.
The Bulletin of the Atomic Scientists has drawn the same connection between AI and biological weapons in its Doomsday Clock statements; see what the Doomsday Clock says about AI, year by year.
Correction, 16 October 2026: An earlier version of this article said the signatories argued that AI systems have already lowered the expertise threshold for designing biological weapons, and attributed to the letter two phrases it does not contain; the letter says the evidence about present-day threats is mixed and that there is a real possibility the knowledge barriers will erode. It also dated the letter to 5 June 2026 rather than 4 June, described its third ask as risk assessments rather than recordkeeping, omitted that the letter also covers synthesis machines and was signed by more than 120 people including the manufacturers of synthetic DNA, attributed the International AI Safety Report 2026 to Yoshua Bengio alone and reported a ranking of risks it does not make, and said that the alignment problem is not on any legislative agenda anywhere. Those passages have been corrected, and the status of S.3741 has been updated.