00:00Lab accidents are actually quite common, more than you would think,
00:02especially as AI is being more utilized in biological research,
00:07that that will potentially run the risk of somebody developing something
00:10in a lab that we haven't seen before.
00:13A shocking suspected pneumonic plague outbreak from a Russian lab
00:17has left biosecurity experts scrambling to figure out what went wrong
00:21and reignited deep concerns about biolab safety.
00:25How secure are biolabs? Who is regulating them?
00:28And could AI help speed up life-saving research
00:31or cause mass destruction in the wrong hands?
00:35I'm J.T. O'Brien. I am the Associate Director of Research
00:38with the Bipartisan Commission on Biodefense at the Atlantic Council.
00:41These labs and the oversight of them depends mostly by the country.
00:46There are international standards for these things, but they're not necessarily enforceable.
00:50So a country like Russia, there's no external entity that will validate
00:55whether or not they're being safe or secure or handling their dangerous research appropriately.
01:00There's actually a global biolabs report that came out in 2023,
01:04and Russia was ranked toward the bottom in terms of their oversight of,
01:09you know, dangerous biological research or biological research in general.
01:12There is the Biological and Toxins and Weapons Convention.
01:15You could kind of think of it as the international organization that is supposed to oversee all of this,
01:20but it is completely voluntary. There is no verification mechanism,
01:24and there has not been one in place since it was formed in the 70s.
01:28There's four different tiers of biological labs.
01:31There's BSL-1 is something that you might do in your high school bio class.
01:35Nothing really dangerous that you're working with,
01:37but as you raise up to BSL level 2, 3, and 4,
01:41that's when things start to get more dangerous when you're working with higher consequence pathogens.
01:45So once you get up to level 4, those are pathogens that are like Ebola
01:50and things that we don't have treatments for.
01:52There's about 51 BSL-4 laboratories in the world,
01:57and there's more in construction, and that spans across 27 different countries.
02:02You know, in the United States and in Western countries,
02:05there are people that go in and oversee that work.
02:07Every single one of those labs has a biosafety officer that oversees that at their entire job.
02:12In Russia, once again, it's a question because we don't know
02:15because we don't have the insight into what's going on on the ground there.
02:19We are kind of in a fog of war, so to speak,
02:22with the situation, and Russia has not been forthcoming.
02:24Lab accidents are actually quite common.
02:26A lot of the times, you just don't hear about it
02:28because there aren't very many consequences.
02:31Sometimes a lab accident can be a needle stick.
02:34It can be a broken test tube.
02:35Sometimes that's just benign.
02:37A lot of worst-case scenarios that keep me up at night with this kind of stuff.
02:41And I think especially as AI is being more utilized in biological research
02:46and is showing a lot of, you know, dual-use capability,
02:49that that will potentially run the risk of somebody developing something
02:52in a lab that we haven't seen before.
02:55And we don't have countermeasures against it.
02:58So think of COVID, but 10 times worse.
03:00And they don't necessarily have to be working with it as a weapon.
03:03They could be legitimately curious.
03:05What if I do this to this virus?
03:07And then you can have a mistake happen.
03:09You know, you can do legitimate research that looks a lot like biological weapons research
03:15or could be turned around and used for biological weapons research.
03:19You know, one example was, and this was a few years ago,
03:22some researchers were using an AI model to find novel chemical compounds for drug discovery.
03:29And all they did was switch the model's tuning to optimize for safety and efficacy of the drug to the
03:38opposite.
03:39And they were able to generate novel chemical weapon formulas.
03:43I think another scenario is, you know, somebody thinking that they're doing something good
03:46and moving too quickly and making a mistake.
03:49And then on the other end of that, you have people who are actually going to try.
03:53And the more powerful model you have, the more capability you have to design something
03:57that is truly scary and horrific in a, you know, WMD catastrophic type of way.
