America Can’t Afford AI Biosecurity Theater

Opinion: New laws would be another win for China at the worst time

Late last week, a high-profile list of frontier AI folks, policy experts and biologists released a letter calling for immediate measures to mitigate the risk of AI-mediated bioterrorism. The details in the letter are scant – you can read it at screendna.org – but the gist is simple. As Dario Amodei, Sam Altman and Demis Hassabis see it, AI is making it too easy for bad actors to harm America using biotechnology, and the solution is to regulate biology rather than the models. In the current reality, where the US biotech industry is getting crushed by China in every metric that matters, introducing shortsighted regulation is not a mistake America can afford to make.

The published letter accompanies a report by the Institute for Progress with specific recommendations on securing the biotech supply chain, along with a NY State Assembly bill proposing new requirements for DNA synthesis service providers. Specific proposals focus on introducing KYC (Know Your Customer) laws as a start, which is a benign request on the surface. The logic here is that bioterrorism threats – like an individual actor trying to revive the extinct but still-deadly smallpox virus or give pathogenicity to an existing bacterial strain – all require designing and manipulating DNA fragments. Biologists have the option of extracting these DNA fragments from live organisms or buying them as a commodity through 200 or so international DNA service providers. Most opt for the latter, but the former still remains an option even with this bill in place.

Although no legal mandate currently exists, the majority of the approximately 50 US DNA synthesis providers proactively screen 90% of incoming orders. Exceptions to this voluntary screening are generally limited to very short sequences and a handful of smaller vendors. The proposed bill adds little actual screening. What the bill would add via the KYC requirements are barriers to American biotech start-ups at the exact moment new ideas and energy are needed in a stumbling industry. Much of the restrictive legislation currently strangling American innovation started out innocuously. Look no further than Good Manufacturing Practice requirements, which have effectively given China the reins of the cell and gene therapy industry. New requirements placed on synthesis providers are likely to go similarly, as there is no way to enforce KYC laws for international vendors, granting them a competitive advantage against our home-grown DNA synthesis companies.

By the authors’ own admission, none of the measures stop a sufficiently driven actor, particularly those operating on behalf of an adversarial state. Offensive knowledge and materials cannot be contained indefinitely, as demonstrated by the benchtop DNA synthesizers already available for purchase. This proposed legislation endorsed by AI leaders is likely to act as a foothold on which costs can be raised, blunting a market that startups rely on. Worse, AI doomers worry models will soon zero-shot novel pathogens. But, if threats can emerge from an effectively limitless sequence space, exhaustive screening is computationally impossible. Codifying existing algorithms is simply not a realistic response to curbing future threats, as they are easy to circumvent as it is. At best, this legislation is ineffective, and at worst it further incentivizes American biotech companies to look abroad for synthesis. None of the legislation applies to Chinese oligo synthesizers, who, along with their AI colleagues, do not red-team, regulate raw AI models or use KYC laws. Chinese counterparts are not thinking much about biosecurity at all, which puts American competitors at a disadvantage if new legislation creates even the slightest friction.

Security must come primarily from defense, but not one built on regulatory grounds. Widespread pathogen surveillance, sequencing, rapid diagnostics and a constant stream of emergent biotechnology platforms act as fantastic countermeasures to any threats. COVID is the clearest example of this. The speed of the mRNA response is owed almost entirely to infrastructure that happened to be available at a time of crisis - mRNA therapies coupled with the selective deregulation of the first Trump administration. Lipid nanoparticle delivery was nascent, but again, was an option simply because the broader innovation environment supported its development. Genomic surveillance was available to help manage the pandemic because of previous work and innovation done by pioneering sequencing providers. A regulation-first posture optimizes against a bad actor that cannot really be stopped, while a capacity-first angle provides an agility best suited to biosecurity challenges.

It’s worth mentioning that these measures are extremely unpopular with lab biologists. A concern from biologists, including exceptionally prominent founders (who are reluctant to speak up because of their business with big AI labs), is that legislation like that proposed and backed by the AI industry acts to further the goals of frontier labs without absorbing any of the blowback. This is all done at the expense of America’s already-struggling bio-economy. While new regulations aimed at the non-existent screening problem are being discussed, frontier labs are fighting pushes from outsiders looking to regulate models over a certain size, citing their heated rivalry with the growing Chinese AI labs. Furthermore, Anthropic has taken to adding strict, self-imposed filters on basic bio-related topics, providing further proof that AI labs buy into the idea of self-governance as opposed to de facto governance.

Frontier labs ultimately do these things because they resist rules that they view as misaligned with the actual pace and shape of their core technology. Just like with those AI-focused bills, it’s likely DNA synthesis restrictions will contribute to strangling the US’s attempts to escape from its current biotech death spiral. It’s a problem best left to industry players, which have already demonstrated a serious commitment to safely providing synthesis services without the need for stifling government intervention.