00:00Now AI is now being used seriously for military purposes.
00:02But it's not some special AI made just for military use.
00:05Even the AI we commonly use has become valuable enough for military applications.
00:08Let's take a proper look at this topic today.
00:09Let's get right into it.
00:11Before that, if you want to get the fastest international news in Korea,
00:13please visit SBS News right away.
00:15I ask for your support.
00:16Now, about Anthropic, the company that developed Claude.
00:18According to reports, Iran tried to use Claude to attack US Navy ships deployed in the Middle East.
00:23According to a report released by Anthropic,
00:25Iran used Claude to gather signal intelligence on ships and aircraft.
00:27They collected military and commercial satellite images to track US warships
00:31and looked for vulnerabilities in the communications systems on board.
00:35Claude did not autonomously carry out physical attacks like launching missiles, though.
00:38Tasks like OSINT analysis, precision targeting, and data refinement,
00:41which used to take military experts weeks to complete,
00:43can now be automated by AI in just a few minutes.
00:45So, from a military perspective, its value for military use has greatly increased.
00:49Anthropic stated that they stopped this threat actor's plan.
00:51In addition, besides Iran, the Houthi rebels also used Claude
00:54to research multistage ballistic missiles with a range of over 2,000 kilometers and
00:57they tried to develop software for guided weapons,
01:00including hypersonic aircraft and other aerial vehicles.
01:02Anthropic is also very concerned about this.
01:04This means that AI is now being actively used in military weapon development and conflict.
01:08Anthropic said that their safety measures blocked some of the requests,
01:11but not all of them, and they also explained this.
01:14So, it's hard to say that we fully understand the scale of AI's military misuse
01:17just based on the currently confirmed cases, isn't it?
01:20In fact, Anthropic also explained that the cases confirmed this time
01:23do not represent the full scale of AI misuse.
01:25Anthropic also said that China and Russia used Claude
01:27and that groups linked to the Chinese government used Claude for Syria
01:30to track and monitor Uyghurs in Syria
01:32and to investigate religious leaders across Asia.
01:34In particular, when it comes to religious leaders across Asia,
01:37they targeted the Catholic cardinals throughout Asia,
01:39the leadership of the Presbyterian Church in Taiwan,
01:41as well as members of Tibetan Buddhism, Buddhist civil society,
01:44and exiles and governments in exile.
01:45Cardinal Yom Sujong and Cardinal Yuhang Sik from our country
01:48were also subjects of investigation.
01:50Also, Russia trained Claude with data from the Ukrainian front.
01:52They tried to build autonomous lethal drone swarms,
01:55and they also exploited it to spread pro-Russian propaganda in African countries.
01:59And it was revealed that a certain country tried to develop deadly biological weapons using Claude,
02:03but was detected and blocked.
02:04However, they did not specify which country it was.
02:06Among the cases disclosed by Anthropic, three were from China.
02:09Two cases were confirmed in Russia,
02:10and one case in Yemen related to the misuse of conventional weapons.
02:14Also, Anthropic emphasized that proper safety measures are...
02:16Without proper safety measures, such capabilities could lead to catastrophic consequences,
02:20so they stated their position that they will develop appropriate safeguards.
02:22In fact, until now, Claude had safety measures in place
02:25to prevent its AI models from being used for malicious hacking,
02:28biological or chemical weapon manufacturing, and so on.
02:30The problem is, it's easier than expected to bypass these measures.
02:33It seems that such cases keep appearing.
02:35Anthropic also said in its report that it hopes the findings will help other developers
02:38detect similar patterns and help governments and civil society
02:40better understand how these new threats are taking shape.
02:42They also said that they hope this will help clearly identify and strengthen a joint defense network.
02:46The value of using AI is truly significant.
02:49AI can make up for the lack of skilled personnel and resources,
02:51which in the past required many intelligence agents.
02:53Now, we've entered an era where one or two people can complete tasks that used to require many.
02:57In the past, weapon development required a lot of experts,
03:00but now, by using Claude Code, it's possible to develop rocket and missile software.
03:03Even if it's not perfect, as long as it can launch to some extent,
03:06it's an extremely valuable asset.
03:08For armed groups, it's now much easier to develop weapons than before.
03:11Also, according to Axios reports,
03:12Researchers receiving government funding reportedly tried to conduct studies at military research institutions
03:17to increase the transmissibility and immune-evasion abilities of the chikungunya virus carried by mosquitoes.
03:21However, Anthropic stated that biological research is for therapeutics.
03:24It can be used for both vaccine development and biological weapon development.
03:27So sometimes it's difficult to distinguish the purpose of the research.
03:30In other words, it means that this can definitely be used for developing biological weapons as well.
03:33Because of this, AI experts are concerned that this kind of AI...
03:36Some say that the pace of AI development should be slowed down.
03:38And there are also people who express concerns about AI being used recklessly and dangerously like this.
03:42There are quite a few people like that.
03:44Of course, aside from that, it seems that this kind of AI development will only accelerate further.
03:47President Trump dismissed concerns that AI could wipe out humanity
03:50and said that falling behind China in the AI race is even more dangerous.
03:53We are in a situation where all efforts are being focused on AI development.
03:56In fact, the speed of AI development is getting faster and faster.
03:58For example, it took GPT-3 to be developed into GPT-4.
04:01It took 1,019 days.
04:02But this pace kept accelerating.
04:04So it took 876 days for GPT-4 to be developed into GPT-5.
04:07And now, from GPT-5.6 to GPT-6, it only took 56 days.
04:11Also, in the case of Claude, it took 444 days to develop from Claude-3 to Claude-4.
04:16But then it only took 383 days to go from Claude-4 to Claude-5.
04:20So the time kept getting shorter.
04:21It only took 84 days to develop from Claude-5 to Claude-5.1.
04:25As you can see, the time is being shortened at an incredibly fast pace.
04:28Right now, GPT-6 Astra is showing incredible performance.
04:31Next month, there could be an AI that shows even greater performance than Astra.
04:35Because of this, AI companies are saying that they need to slow the pace of AI development.
04:39In the case of Amadei from Anthropic, by slowing the pace of development, we can buy another
04:44one or two years before the AI model reaches a dangerously high level of capability.
04:48He said that if we use that time to improve alignment, we can greatly reduce the risk of
04:52serious problems occurring.
04:53He also mentioned that thanks to AI systems that can improve themselves without human
04:57intervention, he also said that after this summer, the development of AI could outpace
05:01our ability to understand and control these systems.
05:03So even if we proceed with development, we must approach it very cautiously.
05:07And it's not just Anthropic saying this.
05:09OpenAI's Altman also expressed agreement with Amadei's opinion on the need for a gradual
05:13approach through X.
05:14Anthropic's chief scientist Evan Hubinger also warned on X after Coxon's resignation announcement
05:19that there is more than a 10% chance AI could wipe out humanity in the next 10 years.
05:23But there's a really important point we need to focus on here.
05:25The biggest change in the military use of AI isn't just that AI can create weapons that
05:29are superior to those made by humans.
05:31What's even more important is that tasks in the military and military intelligence fields,
05:35which used to require massive manpower, expertise, time, and cost, can now likely be carried
05:39out with far fewer people.
05:40In military operations, the process of analyzing information is just as important as obtaining
05:43it.
05:44For example, detecting changes in a specific area by looking at satellite images or comparing data
05:47from multiple sources, it used to require a great deal of expertise to sift through countless
05:51communications and open-source information to extract meaningful content.
05:54If AI is used to analyze such data simultaneously and classify the necessary information, the
05:58burden on people to check all the materials themselves can be greatly reduced.
06:01Especially, the value of using AI is increasing in the process of connecting different types of
06:05information to produce a unified analysis.
06:07This change can also alter the significance of manpower size in the military field.
06:11In the past, preparing for a specific operation required intelligence analysis and software,
06:15where various personnel, such as developers, were once needed, AI can now automate or help
06:19with some of those tasks.
06:20Of course, this doesn't mean that AI will completely replace experts.
06:23Rather, it greatly expands the range of information and tasks that a single expert can handle.
06:27That's more the direction we're headed.
06:28The same change can appear in weapons development as well.
06:30AI may not create complex weapons or weapon systems from scratch, but it can assist with software
06:34development or error analysis.
06:35If AI can be used in development processes like simulation, it can help reduce the time and
06:39manpower needed for development.
06:40Especially when trying to improve new weapons or weapon systems using existing technology,
06:43this new approach can be very beneficial.
06:44These changes are quite significant, but here's where a problem arises.
06:48Military power has long depended on how much capital, manpower, and specialized technology
06:52a country can secure, which has made a big difference.
06:54But if AI comes into this process and reduces the required manpower and time, even smaller
06:58organizations could attempt higher level operations than in the past.
07:01This kind of environment can be created.
07:02Not only countries, but also armed groups and non-state actors can use AI, which is why
07:06the international community is paying attention to this issue.
07:08Right now, with the emergence of drones, we've seen several instances where weaker countries
07:11that would have been overwhelmingly defeated in the past are now able to fight on a more
07:14equal footing.
07:15For example, like the war in Ukraine, or the wars in the Middle East.
07:18But if AI is added to this, even the smallest organizations will be able to carry out military
07:22operations of a certain scale.
07:23In other words, future competition will be about who has more.
07:26It may no longer depend solely on who has more weapons and troops.
07:29Who can use AI more effectively to analyze information, develop software, and support decision-making
07:34will also become an important factor in military power.
07:36It could be, as AI lowers the barriers to building military capability, it may change the capability
07:41gap between countries and organizations.
07:42How this change will affect the future international security environment is something we need to
07:46watch very closely.
07:47Especially since there are so many areas where AI can be used, armed groups could use AI to
07:52devise the most effective methods of terrorism.
07:54That's possible.
07:55That's why we need to prevent someone from abusing AI to create such things.
07:58But I don't think that's going to be as easy as it sounds.
Comments