1 00:00:03,120 --> 00:00:10,040 Speaker 1: Bloomberg Audio Studios, podcasts, radio news. Bloomberg reporter Riley Griffin 2 00:00:10,200 --> 00:00:11,840 Speaker 1: has a pretty unique beat. 3 00:00:12,280 --> 00:00:15,520 Speaker 2: I am a healthcare reporter and I'm based here in Washington, DC, 4 00:00:15,960 --> 00:00:18,840 Speaker 2: and I have a particular interest in where health and 5 00:00:18,960 --> 00:00:20,080 Speaker 2: national security meet. 6 00:00:20,280 --> 00:00:23,480 Speaker 1: And there's been no shortage of stories at that intersection 7 00:00:23,560 --> 00:00:27,000 Speaker 1: of health and national security since the COVID nineteen pandemic. 8 00:00:27,240 --> 00:00:31,840 Speaker 2: We were all incredibly humbled by that moment. And I think, 9 00:00:32,600 --> 00:00:36,839 Speaker 2: you know, prior I had been really focused on the 10 00:00:36,880 --> 00:00:40,080 Speaker 2: pipelines of pharmaceutical companies and how they were going to 11 00:00:40,760 --> 00:00:44,640 Speaker 2: ensure a profitable and booming business. I was not thinking 12 00:00:44,680 --> 00:00:51,720 Speaker 2: about healthcare as something that threatened our economy or threatened 13 00:00:52,120 --> 00:00:54,880 Speaker 2: our ability to act in everyday normal ways. 14 00:00:55,200 --> 00:00:58,080 Speaker 1: Newly alert to the possibility that it could be a threat, 15 00:00:58,480 --> 00:01:01,240 Speaker 1: Riley was intrigued when she got a tip. 16 00:01:01,560 --> 00:01:06,200 Speaker 2: I heard this from government officials who witnessed the briefing 17 00:01:06,760 --> 00:01:08,919 Speaker 2: by a guy who had walked in with a black box. 18 00:01:09,680 --> 00:01:13,880 Speaker 1: A black box, and that briefing wasn't held just anywhere. 19 00:01:14,120 --> 00:01:17,440 Speaker 2: He brought that black box in with him to the 20 00:01:17,480 --> 00:01:22,479 Speaker 2: Eisenhower Executive Office Building EEO B. That's the building right 21 00:01:22,520 --> 00:01:25,039 Speaker 2: next to the West wing. Really, it's where a lot 22 00:01:25,080 --> 00:01:26,319 Speaker 2: of White House staffers are. 23 00:01:26,680 --> 00:01:29,280 Speaker 1: Riley says she talked with several people who were in 24 00:01:29,319 --> 00:01:32,600 Speaker 1: that meeting, and they told her that inside that small 25 00:01:32,680 --> 00:01:34,199 Speaker 1: plastic container. 26 00:01:34,000 --> 00:01:35,959 Speaker 2: Were test tubes with synthetic DNA. 27 00:01:36,360 --> 00:01:39,760 Speaker 1: Synthetic DNA that included ingredients that could be used to 28 00:01:39,800 --> 00:01:44,319 Speaker 1: make deadly diseases. That was frightening enough, but Riley's sources 29 00:01:44,400 --> 00:01:46,880 Speaker 1: told her what really startled them was when they found 30 00:01:46,920 --> 00:01:50,480 Speaker 1: out how that guy, a former UN weapons inspector, had 31 00:01:50,520 --> 00:01:53,760 Speaker 1: figured out what to put into those test tubes. It 32 00:01:53,800 --> 00:01:56,000 Speaker 1: was with the help of an AI chatbot. 33 00:01:56,240 --> 00:01:59,400 Speaker 2: He had his team basically pretending to be a bioterrorist. 34 00:01:59,440 --> 00:02:02,440 Speaker 2: If a biotean art wanted better instructions on how to 35 00:02:02,480 --> 00:02:05,320 Speaker 2: create a biological weapon or help with the process of 36 00:02:05,560 --> 00:02:08,919 Speaker 2: actually making that in a laboratory, what would it ask, 37 00:02:09,000 --> 00:02:09,920 Speaker 2: What would it need to know? 38 00:02:10,200 --> 00:02:13,520 Speaker 1: They typed in questions and they got answers, But it 39 00:02:13,560 --> 00:02:16,840 Speaker 1: wasn't just recipes what to do with that synthetic DNA, 40 00:02:17,480 --> 00:02:20,840 Speaker 1: Riley says. Those government officials heard about how that chatbot 41 00:02:21,040 --> 00:02:24,399 Speaker 1: recommended where to put pathogens and how to deploy them 42 00:02:24,520 --> 00:02:28,200 Speaker 1: to do the most damage, which raised an alarming question. 43 00:02:28,160 --> 00:02:32,040 Speaker 2: If a bad actor could acquire the synthetic DNA and 44 00:02:32,080 --> 00:02:35,000 Speaker 2: the instructions to put it together in a lab, what 45 00:02:35,000 --> 00:02:35,800 Speaker 2: would that mean. 46 00:02:44,639 --> 00:02:46,680 Speaker 1: I'm David Gura and this is the big take from 47 00:02:46,720 --> 00:02:50,200 Speaker 1: Bloomberg News today on the show, how information that could 48 00:02:50,200 --> 00:02:54,160 Speaker 1: be used to make bioweapons slipped through AI safeguards and 49 00:02:54,200 --> 00:02:57,680 Speaker 1: what that tells us about the challenges of regulating is 50 00:02:57,680 --> 00:03:07,640 Speaker 1: still pretty brand new technology. Of course, information about how 51 00:03:07,680 --> 00:03:10,560 Speaker 1: to create deadly weapons of all kinds is out there 52 00:03:10,760 --> 00:03:13,799 Speaker 1: and it has been for a while. Information that could 53 00:03:13,800 --> 00:03:17,960 Speaker 1: help weaponize diseases are buried in books, and Riley Griffin 54 00:03:18,000 --> 00:03:20,560 Speaker 1: says they're online if you know where to look. 55 00:03:20,960 --> 00:03:26,000 Speaker 2: There's a lot of information on the Internet about viruses, pathogens, 56 00:03:26,480 --> 00:03:31,359 Speaker 2: respiratory illnesses, fungal diseases, right, And the reality is there's 57 00:03:31,360 --> 00:03:33,720 Speaker 2: a lot of information on Google that is dangerous when 58 00:03:33,760 --> 00:03:35,840 Speaker 2: it comes to the field of biology. 59 00:03:36,000 --> 00:03:39,240 Speaker 1: I asked Riley, what's different now? What could a bad actor, 60 00:03:39,440 --> 00:03:43,120 Speaker 1: a bioterrorist do with an AI chatbot that he couldn't 61 00:03:43,160 --> 00:03:44,880 Speaker 1: do with a standard search engine. 62 00:03:45,200 --> 00:03:49,840 Speaker 2: One thing that really struck me through the interview process 63 00:03:49,960 --> 00:03:53,800 Speaker 2: was just hearing from some of the researchers explain how good. 64 00:03:54,320 --> 00:03:58,280 Speaker 2: The chatbot was in the brainstorming phase in offering up 65 00:03:58,320 --> 00:04:01,520 Speaker 2: ideas that weren't necessarily in the first place. 66 00:04:01,880 --> 00:04:05,720 Speaker 1: AI companies know that their transformative technology has a lot 67 00:04:05,760 --> 00:04:09,480 Speaker 1: of potential perils, and they're also aware of the damage 68 00:04:09,480 --> 00:04:12,600 Speaker 1: it could do, not just to their business models. So 69 00:04:12,760 --> 00:04:14,600 Speaker 1: many of them have tried to get ahead of a 70 00:04:14,600 --> 00:04:17,640 Speaker 1: lot of this stuff, and Anthropic is one of them. 71 00:04:17,920 --> 00:04:20,600 Speaker 1: It was founded by people who had worked at Open Ai. 72 00:04:21,120 --> 00:04:24,040 Speaker 2: Nanthropic was created really with a bent towards safety. They 73 00:04:24,040 --> 00:04:27,919 Speaker 2: wanted to put in checks and balances, test these things 74 00:04:27,960 --> 00:04:32,560 Speaker 2: out before they reached the public, and commercialize at a 75 00:04:32,600 --> 00:04:34,200 Speaker 2: little bit of a slower pace at the time. 76 00:04:34,400 --> 00:04:37,960 Speaker 1: It was Anthropic who hired that former UN weapons inspector 77 00:04:38,200 --> 00:04:40,960 Speaker 1: who brought that black box to the heart of Washington. 78 00:04:41,600 --> 00:04:46,160 Speaker 1: His name is Rocco Casa Grande, and Anthropic hired him 79 00:04:46,160 --> 00:04:49,600 Speaker 1: to put its AI chatbot, Claude through its paces. 80 00:04:49,920 --> 00:04:52,440 Speaker 2: So that's the origin of this story. He had vested 81 00:04:52,480 --> 00:04:54,120 Speaker 2: it out with a team for more than one hundred 82 00:04:54,160 --> 00:04:59,720 Speaker 2: and fifty hours. He had virologists, microbiologists really examining what 83 00:04:59,800 --> 00:05:01,840 Speaker 2: it is could and couldn't do, and so they went 84 00:05:01,920 --> 00:05:06,159 Speaker 2: through that process and what they found scared Casa Grande. 85 00:05:06,480 --> 00:05:09,200 Speaker 1: He was concerned by how simple it was to get 86 00:05:09,279 --> 00:05:13,160 Speaker 1: pretty accurate instructions for how to make bioweapons and how 87 00:05:13,160 --> 00:05:17,120 Speaker 1: to maximize harm. But Riley says something else made Casa 88 00:05:17,160 --> 00:05:18,839 Speaker 1: Grande even more worried. 89 00:05:19,279 --> 00:05:23,400 Speaker 2: Another notable piece that was shared with me was the 90 00:05:23,480 --> 00:05:26,280 Speaker 2: chatbots could advise on how to acquire the materials, where 91 00:05:26,279 --> 00:05:30,200 Speaker 2: to go purchasing them, and how to evade scrutiny when 92 00:05:30,240 --> 00:05:33,919 Speaker 2: doing so. Some companies that sell the synthetic DNA this 93 00:05:34,000 --> 00:05:37,400 Speaker 2: man made DNA, you know, look to know their customer. 94 00:05:37,480 --> 00:05:39,880 Speaker 2: They want to see who is purchasing this. Is this 95 00:05:39,920 --> 00:05:43,720 Speaker 2: a researcher at the CDC? Is this an academic who 96 00:05:44,440 --> 00:05:48,360 Speaker 2: studies ebola? If not, you know, a red flag is raised. 97 00:05:48,600 --> 00:05:51,599 Speaker 1: But Riley says not all companies that sell man made 98 00:05:51,720 --> 00:05:55,120 Speaker 1: DNA have or follow strict protocols. 99 00:05:55,640 --> 00:05:58,760 Speaker 2: Like with all companies, there's a varying level of commitment 100 00:05:59,000 --> 00:06:03,440 Speaker 2: to that cause, and the chatbok could recommend where to 101 00:06:03,480 --> 00:06:06,400 Speaker 2: look that might be better at evading such scrutiny. 102 00:06:06,680 --> 00:06:10,360 Speaker 1: That motivated Cossa Grande to put together that black box 103 00:06:10,720 --> 00:06:12,520 Speaker 1: and to carry it with him into one of the 104 00:06:12,560 --> 00:06:16,240 Speaker 1: most secure office buildings in the world. He called it 105 00:06:16,279 --> 00:06:19,240 Speaker 1: a stunt, but he wanted to make a point, and 106 00:06:19,360 --> 00:06:23,000 Speaker 1: Riley says it resonated with the people Cossa Grande met with. 107 00:06:23,520 --> 00:06:27,880 Speaker 2: He really struck the imagination of Washington with that black box. 108 00:06:29,160 --> 00:06:32,640 Speaker 1: Coming up after the break, how government officials responded to 109 00:06:32,800 --> 00:06:36,360 Speaker 1: Roco Costa Grande's black box warning and what that could 110 00:06:36,360 --> 00:06:39,880 Speaker 1: mean for everything from AI safety to our ability to 111 00:06:39,960 --> 00:06:53,000 Speaker 1: treat and cure diseases. Just a few months after Roco 112 00:06:53,040 --> 00:06:56,960 Speaker 1: Cassa Grande's briefing, in October of last year, President Joe 113 00:06:57,000 --> 00:07:00,919 Speaker 1: Biden called for greater oversight of government funded research and 114 00:07:01,000 --> 00:07:05,400 Speaker 1: new safety protocols for tech companies and AI developers to. 115 00:07:05,400 --> 00:07:07,960 Speaker 3: Realize the promise of AI and avoid the risk. We 116 00:07:08,040 --> 00:07:12,160 Speaker 3: need to govern this technology, not and there's no other 117 00:07:12,200 --> 00:07:14,520 Speaker 3: way around it. In my view, it must be governed. 118 00:07:14,760 --> 00:07:17,520 Speaker 1: He laid out a new way for AI companies to 119 00:07:17,600 --> 00:07:20,760 Speaker 1: think about the threats posed by the technologies they're creating. 120 00:07:21,080 --> 00:07:24,360 Speaker 3: I'm about to sign an executive order, an executive order 121 00:07:24,760 --> 00:07:28,320 Speaker 3: that is most significant action any government anywhere in the 122 00:07:28,320 --> 00:07:31,480 Speaker 3: world has ever taken on AI safety, security, and trust. 123 00:07:31,800 --> 00:07:35,240 Speaker 1: Well since then, Bloomberg's Riley Griffin says the debate over 124 00:07:35,320 --> 00:07:39,360 Speaker 1: AI regulation in Washington has gotten louder and more urgent 125 00:07:39,800 --> 00:07:41,160 Speaker 1: since October. 126 00:07:41,400 --> 00:07:47,120 Speaker 2: There's really begun in earnest a serious conversation about regulation 127 00:07:47,720 --> 00:07:53,200 Speaker 2: and or what steps can be taken without hard and 128 00:07:53,240 --> 00:07:56,520 Speaker 2: fast rules, just to limit the risks. So this is 129 00:07:56,600 --> 00:08:00,240 Speaker 2: top of mind. Every agency is thinking about it, of 130 00:08:00,240 --> 00:08:04,120 Speaker 2: Homeland Security, Commerce Department, the Department of Defense. I mean, 131 00:08:04,160 --> 00:08:05,960 Speaker 2: it's a full government conversation. 132 00:08:06,360 --> 00:08:09,840 Speaker 1: Late last year, Vice President Kamala Harris announced the establishment 133 00:08:09,960 --> 00:08:14,280 Speaker 1: of the US Artificial Intelligence Safety Institute, and in her remarks, 134 00:08:14,480 --> 00:08:18,880 Speaker 1: Harris specifically singled out AI made bioweapons as a threat. 135 00:08:18,840 --> 00:08:22,760 Speaker 4: From AI enabled cyber attacks at a scale beyond anything 136 00:08:22,760 --> 00:08:28,760 Speaker 4: we've seen before to AI formulated bioweapons that could endanger 137 00:08:28,840 --> 00:08:34,120 Speaker 4: the lives of millions of people. These threats are often 138 00:08:34,160 --> 00:08:39,520 Speaker 4: referred to as the existential threats of AI, because of course, 139 00:08:39,559 --> 00:08:45,520 Speaker 4: they could endanger the very existence of humanity. These threats, 140 00:08:45,800 --> 00:08:51,360 Speaker 4: without question, are profound and they demand global action. 141 00:08:51,800 --> 00:08:54,040 Speaker 1: But at the same time as the government has put 142 00:08:54,120 --> 00:08:58,160 Speaker 1: up new guardrails, some researchers are saying not so fast. 143 00:08:58,520 --> 00:09:02,840 Speaker 2: Nobody wants to stop or certainly many including in the 144 00:09:02,840 --> 00:09:06,400 Speaker 2: scientific world, do not want to stop the innovation that 145 00:09:06,480 --> 00:09:11,520 Speaker 2: AI is contributing to around new treatments and diseases and 146 00:09:11,600 --> 00:09:16,000 Speaker 2: ways of diagnosing healthcare conditions. There are a lot of 147 00:09:16,000 --> 00:09:21,080 Speaker 2: benefits that scientists point to that they fear could be 148 00:09:21,160 --> 00:09:22,760 Speaker 2: hindered by regulation. 149 00:09:23,280 --> 00:09:26,120 Speaker 1: One hundred and seventy four academics have signed a letter 150 00:09:26,280 --> 00:09:30,640 Speaker 1: pledging to use AI in their research responsibly. They argue 151 00:09:30,720 --> 00:09:34,320 Speaker 1: that when it comes to AI, its benefits quote far 152 00:09:34,400 --> 00:09:38,440 Speaker 1: outweigh the potential for harm. Riley says, they're advocating for 153 00:09:38,480 --> 00:09:40,480 Speaker 1: a more measured approach to regulation. 154 00:09:41,040 --> 00:09:44,400 Speaker 2: You know, a focus of the letter is we recognize 155 00:09:44,440 --> 00:09:47,280 Speaker 2: there are some risks, but hold up, we don't want 156 00:09:47,320 --> 00:09:50,719 Speaker 2: to limit the progress that can be achieved. We can 157 00:09:50,760 --> 00:09:54,480 Speaker 2: do good things with these tools, and that is a priority, 158 00:09:54,640 --> 00:09:57,520 Speaker 2: and we promise to move forward with that work while 159 00:09:57,559 --> 00:10:02,720 Speaker 2: practicing safe behaviors around the for example, only purchasing synthetic 160 00:10:02,800 --> 00:10:06,559 Speaker 2: DNA from kind of reliable providers, which isn't again an 161 00:10:06,559 --> 00:10:09,600 Speaker 2: AI question, but there's a it's not just AI. It's 162 00:10:09,640 --> 00:10:13,160 Speaker 2: kind of a broader conversation about security in the field 163 00:10:13,200 --> 00:10:16,240 Speaker 2: of biology. As technology advances. 164 00:10:15,960 --> 00:10:20,319 Speaker 1: Some critics decry what's being called AI doomerism. What they 165 00:10:20,400 --> 00:10:24,440 Speaker 1: see as a misplaced worry that generative AI poses an 166 00:10:24,520 --> 00:10:26,359 Speaker 1: existential threat to humanity. 167 00:10:26,600 --> 00:10:29,960 Speaker 2: Some people think the conversation about biological threats and AI 168 00:10:30,080 --> 00:10:30,400 Speaker 2: is a. 169 00:10:30,320 --> 00:10:33,320 Speaker 1: Distraction, and Riley notes that while Casa Grande and his 170 00:10:33,400 --> 00:10:35,880 Speaker 1: team were able to do something with that chatbot that 171 00:10:36,040 --> 00:10:39,520 Speaker 1: is objectively scary, the test they ran led to a 172 00:10:39,559 --> 00:10:40,440 Speaker 1: positive outcome. 173 00:10:40,600 --> 00:10:44,000 Speaker 2: My reporting suggests that these initial briefings were really the 174 00:10:44,000 --> 00:10:49,640 Speaker 2: beginning point of a massive undertaking by government to think 175 00:10:49,679 --> 00:10:52,520 Speaker 2: about biological risk in a new way in the wake 176 00:10:52,559 --> 00:10:55,480 Speaker 2: of a pandemic, but also in the wake of a 177 00:10:55,480 --> 00:10:57,520 Speaker 2: fast moving emerging technology. 178 00:10:57,720 --> 00:11:01,000 Speaker 1: And those meetings also led to calls for a broader 179 00:11:01,040 --> 00:11:04,760 Speaker 1: conversation so that it's not just the largest AI companies 180 00:11:04,920 --> 00:11:08,559 Speaker 1: like Anthropic and Open Ai that are briefing government officials. 181 00:11:08,960 --> 00:11:12,280 Speaker 2: You know, one interesting comment that I've heard is many 182 00:11:12,360 --> 00:11:15,880 Speaker 2: would like to see smaller AI companies more present in 183 00:11:15,920 --> 00:11:19,680 Speaker 2: the room. It's the biggest players that are engaged with 184 00:11:19,720 --> 00:11:22,480 Speaker 2: government on this subject, but some of the smaller companies, 185 00:11:22,520 --> 00:11:27,680 Speaker 2: particularly those tailored to biological data, may not have as 186 00:11:27,800 --> 00:11:28,920 Speaker 2: active of a voice. 187 00:11:29,080 --> 00:11:33,080 Speaker 1: The debate continues, and according to Riley, the trend is 188 00:11:33,160 --> 00:11:35,960 Speaker 1: looking better than it was last year when she got 189 00:11:35,960 --> 00:11:38,000 Speaker 1: that tip about that black box. 190 00:11:38,360 --> 00:11:41,080 Speaker 2: Hopefully we're at a place where it's harder to get 191 00:11:41,120 --> 00:11:43,079 Speaker 2: that information from the chatbots. 192 00:11:43,240 --> 00:11:46,720 Speaker 1: One of Anthropics founders told Riley the company has made 193 00:11:46,840 --> 00:11:50,800 Speaker 1: changes to address vulnerabilities Cosa Grande and his team identified, 194 00:11:51,360 --> 00:11:53,120 Speaker 1: but more work is needed. 195 00:11:53,760 --> 00:11:53,880 Speaker 4: Well. 196 00:11:53,960 --> 00:11:56,400 Speaker 1: Riley says she doesn't want the takeaway from this story 197 00:11:56,520 --> 00:11:59,200 Speaker 1: to be one of fear, but more so about how 198 00:11:59,200 --> 00:12:01,520 Speaker 1: we think about what we do when there is so 199 00:12:01,600 --> 00:12:04,400 Speaker 1: much innovation happening in so many places. 200 00:12:05,000 --> 00:12:06,880 Speaker 2: You know, what I want the takeaway from the story 201 00:12:06,880 --> 00:12:09,160 Speaker 2: to be is not that we are in a doomsday 202 00:12:09,160 --> 00:12:13,120 Speaker 2: scenario where the worst is here and present and going 203 00:12:13,160 --> 00:12:17,120 Speaker 2: to impact us tomorrow. It's more around the nuance of 204 00:12:17,160 --> 00:12:21,079 Speaker 2: a debate about how to regulate a fast moving technology. 205 00:12:21,720 --> 00:12:25,000 Speaker 2: And I think everybody knows that there's a question around 206 00:12:25,160 --> 00:12:30,040 Speaker 2: regulation of AI because AI is accessible, but the field 207 00:12:30,080 --> 00:12:34,720 Speaker 2: of biology is not so present in that conversation, and 208 00:12:34,760 --> 00:12:39,240 Speaker 2: we talk about synthetic biology or this kind of revolution 209 00:12:39,640 --> 00:12:42,240 Speaker 2: in the field of biology. It is moving just as quickly, 210 00:12:42,520 --> 00:12:44,400 Speaker 2: so that will be kind of my note is what 211 00:12:44,559 --> 00:12:50,600 Speaker 2: happens when too fast moving revolutionary technologies crash into each other. 212 00:12:56,920 --> 00:12:59,720 Speaker 1: This is the Big Take from Bloomberg News. I'm David Gera. 213 00:13:00,320 --> 00:13:03,120 Speaker 1: This episode was produced by Thomas lou It was fact 214 00:13:03,200 --> 00:13:06,359 Speaker 1: checked by Alex Segura. It was mixed by Robert Williams. 215 00:13:06,880 --> 00:13:10,120 Speaker 1: Our senior producers are Naomi Shaven and Kim Gittleson, who 216 00:13:10,160 --> 00:13:13,920 Speaker 1: also edited this episode along with Becca Greenfield. Our senior 217 00:13:14,000 --> 00:13:17,880 Speaker 1: editor is Elizabeth Ponso, Nicole Beemster bor is. Our executive producer. 218 00:13:18,080 --> 00:13:21,679 Speaker 1: Sage Bauman is Bloomberg's head of podcasts. Thanks so much 219 00:13:21,679 --> 00:13:24,560 Speaker 1: for listening. Please follow and review The Big Take wherever 220 00:13:24,600 --> 00:13:27,600 Speaker 1: you get your podcasts. It helps new listeners find the show. 221 00:13:28,120 --> 00:13:33,200 Speaker 1: We'll be back on Monday.