Companies that sell made-to-order DNA operate much like any other industrial supplier. A researcher, a biotech company, or, in theory, anyone else submits an order specifying a genetic sequence, and the supplier prints it and ships it. The process is largely automated, the cost has fallen sharply over the past decade, and the customer base is global.
The CEOs of OpenAI, Anthropic, Google DeepMind, and Microsoft AI signed a public letter on June 3 asking Congress to make screening of these orders legally required. The letter argues that AI tools are eroding what used to be a natural barrier: the sheer difficulty of knowing enough biology to design something dangerous. That barrier is getting lower.
The AI CEOs are not exaggerating about the technical problem. A Microsoft research team spent two years quietly testing whether AI protein-design tools could create variants of known toxins, such as ricin, that would slip past the filters DNA synthesis companies use. The results, published in the journal Science in October 2025, were clear: one widely used screening tool missed more than 75 percent of the AI-designed variants. The sequences had been rewritten at the code level while preserving the toxic structure, making them look harmless to the software.
This is the core issue: current screening tools work by comparing an order against a library of known dangerous sequences. AI can produce a sequence that is functionally similar to something dangerous but looks different enough to pass the check. It is the biological equivalent of changing enough words in a document to fool a plagiarism detector while keeping the meaning intact.
Two separate bills are moving through Congress. The first, H.R. 3029, passed the House Science Committee in April 2025 but focuses on voluntary standards rather than hard requirements. The second, introduced by Senators Tom Cotton and Amy Klobuchar in February 2026, would make screening and customer verification mandatory for all gene synthesis providers operating in the United States, with the Secretary of Commerce responsible for enforcement. The AI CEOs' letter is effectively a push to move the needle from voluntary to mandatory.
The letter calls for two specific measures: every order screened against a database of dangerous sequences, and records kept so that investigators can trace suspicious purchasing patterns, including cases where someone splits an order across multiple suppliers to avoid triggering alarms individually.
The screening problem also has a gap at the short-sequence level. DNA fragments below a certain length are often not screened at all, yet standard laboratory techniques can assemble those short fragments into something dangerous. A 2025 RAND report identified this as one of the key weak points in the biological supply chain.
For business operators, there are a few things worth watching. If the Senate bill passes, companies that receive US federal funding will only be allowed to purchase synthetic genetic material from fully compliant suppliers. That requirement already exists in a limited form under Biden-era guidelines, but the new bill would expand and enforce it. Smaller DNA synthesis vendors may struggle with the compliance cost, which could consolidate the market toward larger players who already have the infrastructure. Companies that build biosecurity screening tools stand to gain directly if a federal mandate creates steady demand for their products.
The broader signal here is structural. AI risk has been a software conversation for the past few years. This letter marks a moment when that conversation formally crosses into physical supply chains, procurement rules, and industrial regulation. Whatever Congress does with these bills, the underlying dynamic does not reverse: AI tools will keep getting better at biology, and the gap between "knowing how" and "being able to do" will keep narrowing.