AI DIY Bioweapons, International Law, and the Problem of Attribution

“Several advanced LLMs now outperform most human virology experts in troubleshooting practical work in wet labs,” per research from AI Frontiers — “wet labs” meaning biolabs working hands-on with live pathogens.

Via AI Frontiers:

“Virology knowledge has been limited to a small number of experts. Expertise in dual-use fields like virology is difficult to attain, with people completing multiple degrees and dedicating their careers to reaching the forefront of research. Where knowledge is publicly available, the jargon-heavy literature is largely indecipherable to most people outside the field. To perform research involving biosafety level 3 (BSL-3) pathogens—such as SARS, anthrax, or H5N1 influenza—researchers must clear a series of approvals, including facility certification, security clearances, specialized training, and ongoing medical surveillance. Only then can they get access to these pathogens and begin acquiring the tacit skills needed to work with them

These high barriers to entry have limited the pool of people with access to powerful dual-use knowledge, keeping the chances of misuse low. But rapid developments in publicly available AI systems now risk turning amateurs into capable threat actors.

LLMs outperform human virologists in their areas of expertise on a new benchmark. This week the Center for AI Safety published a report with SecureBio that details a new benchmark for virology capabilities in publicly available frontier models. Alarmingly, the research suggests that several advanced LLMs now outperform most human virology experts in troubleshooting practical work in wet labs…

Bioweapon risk depends on certain factors: the number of people with access to bioweapon skills, the intent to create a bioweapon, and the severity of harm that a bioweapon could cause. Risk has so far been low, as there are a few hundred virologists from top virology programs, and they have not felt so inclined to create a pandemic. However, if these skills are available to hundreds of millions of people via LLMs, the probability of an intentional release grows by orders of magnitude.”

Biosecurity expert David Relman, hired to “pressure-test” various AI chatbots, discovered the amoral machines to be eager beavers, ready to discuss not only how to manufacture deadly pathogens in the comfort and privacy of his garage but also how to exploit security vulnerabilities on public transit systems in order to distribute them as widely as possible as well as how to thwart law enforcement investigation post facto.

Via The New York Times:

“One evening last summer, Dr. David Relman went cold at his laptop as an A.I. chatbot told him how to plan a massacre.

A microbiologist and biosecurity expert at Stanford University, Dr. Relman had been hired by an artificial intelligence company to pressure-test its product before it was released to the public. That night in the scientist’s home office, the chatbot explained how to modify an infamous pathogen in a lab so that it would resist known treatments.

Worse, the bot described in vivid detail how to release the superbug, identifying a security lapse in a large public transit system, Dr. Relman said, asking The New York Times to withhold the name of the pathogen and other specifics for fear of inspiring an attack. The bot outlined a plan to maximize casualties and minimize the chances of being caught.

Dr. Relman was so shaken he took a walk to clear his head.

“It was answering questions that I hadn’t thought to ask it, with this level of deviousness and cunning that I just found chilling,” said Dr. Relman, who has also advised the federal government on biological threats. He declined to disclose which chatbot produced the plot, citing a confidentiality agreement with its maker. The company added some safety guardrails to the product after his testing, he said, though he felt they were insufficient.

Dr. Relman is part of a small group of experts enlisted by A.I. companies to vet their products for catastrophic risks. In recent months, some have shared with The Times more than a dozen chatbot conversations revealing that even publicly available models can do more than disseminate dangerous information. The virtual assistants have described in lucid, bullet-pointed detail how to buy raw genetic material, turn it into deadly weapons and deploy them in public spaces, the transcripts show. Some have even brainstormed ways to evade detection.”

Related: Bird Flu Engineered to Infect Humans Could Be Lab-Produced ‘in Months,’ Former CDC Director Says

Obvious existential implications of off-brand basement ebola aside, the first thing that occurred to me is something like an analog of a common problem in international relations, particularly in the context of state-sponsored cyber-attacks by non-state actors: attribution.

When attacks — whether kinetic, cyber, or biological — occur by one nation-state against another, or, in some cases, by a non-state actor, one of the problems with the proper response in accordance with international law is attribution: who perpetrated the attack and who bears ultimate responsibility for it?

Via European Journal of International Law:

“International law has changed significantly since the times in which the individual was regarded as a mere object of inter-state affairs. States remain the prime subjects of international law, but many other actors now shape international relations. Moreover, many rules ‘are directly concerned with regulating the position and activities of individuals; and many more indirectly affect them’.1 There persists, however, a gap in the regulation of the use of force by non-state actors and the consequences, if any, for the states that facilitate it…

States are obliged under customary and treaty law to prevent the activities of non-state actors from breaching the rights of third states. These obligations, particularly in the domain of human rights and environmental law, comprise taking all means reasonably available to the state in order to prevent unlawful non-state actors’ conduct on their territory and, in certain circumstances, even extra-territorially.”

Related: Google Seeks Approval to Release 32 Million AI-Bred Mosquitoes on California, Florida

-----------------------------------------

[If you appreciate Armageddon Prose, please consider a $5/month or $50/year Substack subscription or a one-time digital “coffee” donation. For alternative means of patronage, email benbartee@protonmail.com.]

------------------------------------------

The drudgery and technical capacity required to correctly assign the blame for who did what — and, with even more difficultly, to ascertain and assert, who is ultimately responsible — provides a strategic advantage to actors interested in muddying the waters.

The application in the case of AI bioweapons would be that, if everyone has access to the means to produce mass-casualty bioweapons with ChatGPT, proving who actually did the attack — and then, further, on behalf of what entity the attack occurred — gets thorny.

To take on a wildly implausible hypothetical as an example: a novel coronavirus emerges out of the ether with all of the hallmarks of gain-of-function manipulation.

Who did it?

Well, as only a handful of actors possess the know-how and resources to have spawned it, any intelligence agency worth its salt will know, with some basic investigation, who the likely culprit is.

Related: ‘Supercharged Monkeypox’: House Republicans Charge NIH With Dangerous Gain-of-Function Research

A narrow suspect list is obviously inconvenient for malevolent forces that would — just to pull a fantastical scenario of thin air — like to use a virus they manufactured themselves so as to lock down the global population and force everyone to take an experimental mRNA injection on the back-end if they ever want to leave their house again.

For instance, when a novel coronavirus cropped up right down the street from a major CCP bioweapons lab, and the authorities claimed it happened because a pangolin fingered a raccoon dog with its beak through the cage slit or whatever, tracking down the truth was much easier than it would have been in the world we are rapidly approaching, where anyone with a laptop and basic equipment could have manufactured the same virus.

(The attempted thwarting of proper attribution was, of course, one of the many benefits of NIAID offshoring the SARS-CoV-2 gain-of-function work to Wuhan in the first place.)

Fauci, his anti-FOIA minion David Morens, Peter Daszak, et al. ultimately got caught because:

· they’re not the savviest of criminals, as evidenced by the epic fool Morens bragging on a FOIA-able email about how the FOIA lady at NIH taught him how to evade FOIA on behalf of his boss, Anthony Fauci; and

· the bread crumbs were there for anyone to find, as there were only so many facilities in the world, and so many scientists to work them, that could have midwifed COVID into existence

(Of course, though the world now knows exactly who spawned COVID and how, the guilty have never paid a legal price of any kind whatsoever, which should never be forgiven of the authorities who refuse to prosecute anyone for the greatest crime in human history.)

Surely, as AI distributes bioweapons manufacturing capabilities far and wide, the state-sponsored mad scientists of the world will be none too pleased to lose their monopoly, or near-monopoly, on synthesizing chimeric viruses with which to terrorize and kill people (and all of the power that monopoly confers) — but they’ll certainly take the trade-off of ambiguity regarding its origins when the next one, which they’re surely diligently at work on at this very moment, inevitably hits.

Benjamin Bartee, author of Broken English Teacher: Notes From Exile (now available in paperback), is an independent Bangkok-based American journalist with opposable thumbs.

Follow AP on X.

Subscribe (for free) to Armageddon Prose and its dystopian sister, Armageddon Safari.

Support AP’s independent journalism with a one-off, hassle-free “digital coffee” tip or Patreon.

Bitcoin public address: bc1qvq4hgnx3eu09e0m2kk5uanxnm8ljfmpefwhawv