1 hour ago
27
The United States has spent decades treating biosecurity as a slow-moving policy problem. Artificial intelligence just made it a fast-moving one.
Gautam Mukunda, writing for Bloomberg on September 17, 2026, argues that AI has fundamentally changed the calculus around biological risk. His core demand is straightforward: make screening of synthesized DNA orders mandatory, not optional.
AI tools now allow people without advanced biology training to design genetic sequences capable of producing harmful pathogens and toxins.
Why voluntary guidelines are not cutting it
For years, the US biosecurity framework around DNA synthesis has relied on companies voluntarily screening orders against databases of known dangerous sequences.
A Microsoft-led study released in October 2025 found that AI-generated protein variants could design toxins capable of evading commercial DNA screening tools entirely. Detection rates for certain AI-designed sequences fell to as low as 32%, meaning roughly two in three dangerous designs slipped through undetected.
Industry leaders are already on board
In June 2026, an open letter signed by leaders from prominent AI companies, including OpenAI CEO Sam Altman and Google DeepMind CEO Demis Hassabis, urged Congress to implement stricter regulations governing DNA and RNA synthesis.
Congress has also begun moving, at least on paper. The Biosecurity Modernization and Innovation Act of 2026 reflects growing legislative acknowledgment that the existing voluntary framework needs structural reinforcement.
The national security community is taking notice
A survey cited in the research found that 70% of people in the national security field believe AI substantially heightens bioweapon risks.
Mandatory DNA synthesis screening would create a checkpoint at a specific, identifiable chokepoint: the moment when someone orders genetic material for manufacture. A mandatory screening requirement would mean every order gets checked against threat databases before it ships, rather than relying on each company to decide independently whether to screen.
If AI can already produce toxin designs that evade 68% of current screening attempts, the screening databases themselves need to evolve alongside AI capabilities, not just exist as a static archive of yesterday’s threats.
The 70% figure from national security practitioners, the cross-industry letter from AI CEOs, and the Microsoft detection-rate data all point toward the same conclusion: the biosecurity community has reached rough consensus on the problem. The remaining question is whether legislative action arrives before that detection gap becomes something worse than a statistic.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·