AI industry unprepared for potential catastrophe, says Gautam Mukunda
Yale lecturer warns that competitive pressures push AI companies toward growth over safety, with biological threats topping the risk list
Gautam Mukunda, a Bloomberg Opinion columnist and lecturer at the Yale School of Management, has been sounding an increasingly urgent alarm: the AI industry’s approach to catastrophic risk is fundamentally broken, and the consequences could be biological in nature.
The biological threat no one wants to talk about
Mukunda has zeroed in on the intersection of AI and biology, specifically the ability of AI systems to generate DNA sequences that could be used to create dangerous pathogens.
A Microsoft study from October 2025 found that AI-generated DNA sequences could escape detection by commercial screening tools with as low as a 32% success rate. Put another way: roughly two-thirds of the time, existing tools failed to flag potentially dangerous synthetic DNA created with AI assistance.
Current screening for synthesized DNA is largely voluntary. Companies that manufacture synthetic DNA can choose whether or not to screen orders against databases of known dangerous sequences. Mukunda has argued repeatedly that this opt-in system creates obvious gaps, especially as AI makes it easier to design novel sequences that do not match anything in existing threat databases.
Self-regulation is not cutting it
In June 2026, a group of prominent AI leaders, including Sam Altman of OpenAI and Demis Hassabis of Google DeepMind, signed an open letter urging Congress to impose mandatory screening measures for DNA synthesis.
During an appearance on October 4, 2026, Mukunda asserted that voluntary AI safety measures are inadequate, stating that competition drives companies towards growth over safety.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
In an August 2026 Bloomberg segment, Mukunda voiced concerns over public mistrust in Big Tech’s self-governance regarding AI risks.
The intelligence overestimation problem
In a column published on September 24, 2026, Mukunda critiqued the AI industry for overestimating what intelligence alone can accomplish while ignoring the physical experimentation bottlenecks that constrain real-world applications. Mukunda’s counterpoint is that plenty of scientific and engineering challenges require actual physical experimentation, processes that take time regardless of how smart the system directing them might be.
This misunderstanding, Mukunda argues, creates a dangerous blind spot. If leaders believe AI can simply think its way out of any problem, they may underinvest in the physical safeguards, like robust DNA screening infrastructure, that are needed to prevent misuse.
What this means for the AI landscape
Anthropic, one of the major AI safety-focused companies, has been navigating plans for an IPO while simultaneously facing internal scrutiny over its own safety practices.
The US has not yet passed comprehensive federal legislation on AI safety. Mandatory DNA synthesis screening of the kind Mukunda and the open letter signatories have called for would represent a concrete, enforceable step, but it requires Congressional action.