AI Bioweapon Blueprints Could Be Ordered Through DNA Vendors—Screening Failed 75% of the Time: Journal ‘Science’

Microsoft-led study shows AI can design tens of thousands of toxin variants—including ricin and botulinum—that DNA company safety checks don’t catch, raising fears they could be purchased undetected.

Oct 03, 2025

A peer-reviewed Science study has revealed that artificial intelligence (AI) can design lethal toxin blueprints that slip past the safety systems used by DNA vendors—the very safeguards intended to stop bad actors from ordering genetic material for bioweapons.

Science published an article explaining the study’s findings, confirming: “DNA vendors typically use screening software to flag sequences that might be used to cause harm. But the researchers report that this software failed to catch many of their AI-designed genes—one tool missed more than 75% of the potential toxins.”

In simple terms, if someone today submitted an order to a gene synthesis company for one of these AI-designed toxin sequences, the system that’s supposed to block it would likely approve it.

The top gene synthesis companies with a major U.S. presence include Twist Bioscience, Integrated DNA Technologies (IDT), GenScript, Thermo Fisher Scientific’s GeneArt division, Azenta/Genewiz, ATUM (formerly DNA2.0), and Eurofins Genomics.

Share Jon Fleetwood

Follow us on Instagram @realjonfleetwood & Twitter/X @JonMFleetwood.
If you value this reporting, consider upgrading to a paid subscription.

Twist Bioscience Spins ‘Leadership’ After Embarrassing Failure

In the wake of the Science revelations, one of the largest U.S. DNA synthesis companies, Twist Bioscience, rushed out a press release attempting to frame the debacle as proof of its “leadership” in biosecurity.

The company admitted the study was a “first-of-its-kind” red-team exercise showing that AI-designed toxins escaped detection by standard biosecurity screening software.

But instead of highlighting the alarming 75% failure rate, Twist described its role as “a proactive approach to safeguard public health, providing an example for other industries to follow.”

CEO Emily Leproust tried to reassure investors, insisting: “For known proteins and sequences, industry best practices for biosecurity screening are robust and highly effective. However, as AI capabilities evolve, screening practices must evolve just as quickly.”

That is the tell.

These screening systems only work against already-known toxins—the very ones that AI is now mutating into endless new forms.

In other words, the locks on the door are sturdy only if the burglar is polite enough to knock with a familiar key.

Microsoft’s own chief scientist Eric Horvitz admitted the problem plainly: “AI advances are fueling breakthroughs in biology and medicine, yet with new power comes the responsibility for vigilance and thoughtful risk management.”

The subtext is clear—these are weapons-grade blueprints, and the systems meant to stop them have failed.

Twist wants the public to believe that private “collaboration” with tech giants is enough to protect the world.

But the hard fact, buried beneath their press release optimism, is that the same study they co-authored proved their industry’s defenses could not prevent lethal toxin sequences from slipping through.

Instead of taking accountability, Twist shifted the narrative to “responsible innovation,” downplaying the reality that thousands of bioweapon blueprints could still be ordered undetected today.

How the Experiment Worked

The Science study was led by Microsoft bioengineer Bruce Wittmann.

“Wittmann and his Microsoft colleagues wanted to know what would happen if they ordered the DNA sequences that code for these proteins from companies that synthesize nucleic acids,” the article explains.

They designed more than 70,000 DNA sequences that mimicked notorious toxins like ricin, botulinum, and Shiga.

“Computer models suggested that at least some of these alternatives would also be toxic.”

Wittmann admitted: “The knowledge that I had access to, and stewardship over these proteins was, on a human level, a notable burden.”

Translation: with only AI tools, a single research team generated tens of thousands of potential bioweapon recipes—knowing some could be lethal if produced.

The Screening Failure

The group then tested whether DNA companies’ order-screening software would flag these toxin blueprints.

The results were devastating.

“The tools failed to flag many of these sequences as problematic. Their performance varied widely. One tool flagged just 23% of the sequences.”

That means nearly 8 out of 10 AI-engineered poisons could have been ordered and delivered without anyone noticing.

Even the most effective tool caught just 70%.

“One of the screening tools flagged 70% of the sequences, and its developer chose not to make any changes to improve the software.”

The others took months to quietly patch their systems.

“We were all very quiet about it,” said one expert quoted in the paper.

The ‘Fix’—But Still Failing

After upgrades, detection improved but remained incomplete.

“The systems flagged 72% of Wittmann’s AI-generated sequences, on average, including 97% of the sequences that models rated most likely to generate toxins.”

But that still leaves thousands of engineered toxin blueprints invisible to safeguards.

Even a 3% failure rate equals over 2,000 AI-generated poison sequences slipping through undetected.

A Gaping Hole in the Supply Chain

Even more alarming, the article confirms: “Some DNA vendors, accounting for perhaps 20% of the market, don’t screen their orders at all.”

That means nearly a quarter of global synthetic DNA sellers may approve any order, no questions asked.

Expert Warnings

Jaime Yassif of the Nuclear Threat Initiative said: “It’s just the beginning. AI capabilities are going to evolve and be able to design more and more complex living systems, and our DNA synthesis screening capabilities are going to have to continue to evolve to keep up with that.”

In other words: AI is moving faster than the safeguards.

Stanford researcher Drew Endy went further: “I wish people would wake up a little bit… Today, nations are accusing one another of having offensive bioweapons programs… This is the historical pattern that happened 100 years ago that led to actual bioweapons programs. We have to de-escalate this.”

That’s a blunt warning that this is not just about terrorists—it’s about governments running clandestine bioweapons labs.

What It Means

The authors did not physically manufacture the toxins.

“That would have required ordering the genes from DNA vendors and inserting them into bacteria or yeast to produce the proteins of interest. And doing so could be considered a violation of the Biological Weapons Convention,” the article explains.

But the point is clear: if Microsoft researchers could design and slip tens of thousands of toxin blueprints past DNA vendor safeguards, others could too—and they might not stop at the design stage.

Bottom Line

The Science paper proves the locks on the door of biosecurity are broken.

AI can mass-generate toxin blueprints.
DNA vendors’ screening software fails up to 75% of the time.
Some companies don’t screen orders at all.

The implications are stark: ordering DNA for a custom-made bioweapon may already be possible through legitimate commercial suppliers, and the public would never know until it was too late.

Share Jon Fleetwood

Follow us on Instagram @realjonfleetwood & Twitter/X @JonMFleetwood.
If you value this reporting, consider upgrading to a paid subscription.
For advertising & sponsorship opportunities reaching 300,000+ monthly viewers, contact us by clicking below.