Biosecurity risk from AI-designed biology
- Seoyoung Kang
- 5 days ago
- 1 min read
Oct 13, 2025
Seoyoung Kang
Recently, due to an increase in AI usage, not only daily life but also biosecurity is getting threatened. Just as a new virus mutation that avoids vaccines has emerged, protein variants created by AI can avoid the existing security system.
A Microsoft research team tried to create tons of millions of variants of dangerous proteins (toxin) with AI protein design tools, and the problem is that the existing DNA order screening system cannot detect it properly. About 3% of some variants are still not detected even if the screening update (patch) is applied.
The discovery is being accepted as a significant loophole in the field of AI + biosecurity, which is called a kind of "zero-day" issue.
In this passage, "Zero Day" refers to a situation where new, modified proteins designed by AI are suddenly discovered without anyone knowing the problem of bypassing the existing security filter.
The combination of AI and biotechnology tools declines the technical barrier a lot.
In the past, designing biological weapons required a lot of expertise and laboratory experience, but as AI provides design tools, code (sequence) generation, rescue prediction, etc., are automated, and the entry barrier is being lowered.
Therefore, it is important to review the possibility of behavioral testing or functional validation in the laboratory for sequence variations generated by AI tools.
