00:00Bloomberg's learned that Anthropic is accusing Alibaba of using thousands of fake accounts to
00:04illicitly access its clawed AI model. The US AI developer says rivals, including Alibaba,
00:09may be trying to build chatbots cheaply through a technique known as adversarial distillation.
00:14Tech and industrial policy reporter Maggie Eastland joins us now from Washington.
00:19Talk us through what exactly the accusations are and I'm also curious as to what
00:25adversarial distillation comprises. Great question. So Anthropic is accusing Alibaba,
00:33one of its biggest Chinese competitors, of essentially using thousands of fraudulent
00:38accounts to access clawed. What's important here is Anthropic actually doesn't allow Chinese users
00:44to tap any of its models, even as regular users. So somehow, Anthropic is alleging that Alibaba
00:51got access and then queried the model, asking it specific questions about agentic reasoning,
00:58coding capabilities, and then using those outputs to feed into its own AI model. Now,
01:04Anthropic alleges that this is essentially a violation of the intellectual property that
01:10they've created and all the investment that they've put into making clawed.
01:15We know that Bababa's already got its own issues. It's litigating against the Department of Defense
01:19when it comes to this Pentagon blacklist as well. Have we seen an additional response from Washington
01:24when it comes to these continued warnings and tensions?
01:29There are early signs in Congress that lawmakers are interested in this issue,
01:35which as you said, the company defines as adversarial distillation. So in both the House
01:40and the Senate, a bipartisan group has formed and they're introducing and pushing for legislation
01:45that would direct the administration, specifically the Commerce Department,
01:50to consider blacklisting entities that are found to be distilling or taking data from US model companies.
02:00So that, if that follows through, that could be a big hit to some of these big Chinese AI firms.
02:08Why is this issue so crucial for Anthropic at this point?
02:14I think Anthropic has a lot of financial interest right now in this issue. The company is valued by
02:21private investors at about $965 billion, obviously a huge valuation and they're looking to IPO very soon
02:28this year. So clearly they have some interest in assuring investors how they're dealing with some of
02:37these cheap competing models from China. And they're also concerned about about the national security
02:44here, of course. And they mentioned that in this letter that we we've obtained. But, you know,
02:48I'd be remiss not to mention that they have a clear financial interest in sort of tackling this issue of
02:54cheaper AI models from China that are being widely used, including in the US.
Comments