Anthropic Says Alibaba Must Be Punished For Largest Claude Cloning Attack
- Reference: 0184087522
- News link: https://yro.slashdot.org/story/26/06/25/1810226/anthropic-says-alibaba-must-be-punished-for-largest-claude-cloning-attack
- Source link:
> Anthropic has accused the Chinese firm Alibaba of [1]launching the largest attack yet attempting to clone Claude , as China races to match the capabilities of Anthropic's leading model following [2]Mythos' release and [3]subsequent restriction from foreign markets. Ars obtained a June 10 letter sent to Senators Tim Scott (R-S.C.) and Elizabeth Warren (D-Mass.) one day ahead of a Senate committee hearing on "AI and the American Dream." In the letter, Anthropic shared "new, confidential evidence of the largest campaign to illicitly extract Claude's capabilities we have ever measured."
>
> The attacks occurred between April 22 and June 5, when "operators afliated with Alibaba and Alibaba Qwen, Alibaba's AI lab" allegedly generated "more than 28.8 million exchanges with Claude through almost 25,000 fraudulent accounts," Anthropic said. Violating Claude's terms of service and access restrictions, this campaign "targeted some of Claude's most valuable capabilities, such as agentic reasoning, software engineering, and long-horizon tasks." According to Anthropic, Alibaba evaded detection by "using obfuscation techniques and proxy networks." As Chinese demand for reliable obfuscation techniques increases, Anthropic warned there's already "a growing circumvention economy" to fuel an ever-expanding web of future distillation attacks. [...]
>
> "Alibaba is governed by an independent board, none of whom has any military affiliation," Alibaba said. "Its products and services are built for retail, logistics, and enterprise information technology -- not weapons, defense, or intelligence." Anthropic appears unconvinced, however, that Alibaba isn't working with the Chinese government. In the letter, Anthropic warned that without stronger interventions, these distillation attacks will "help China reach Mythos Preview-level capabilities sooner."
>
> To keep the US ahead of China, Anthropic recommended that Congress pass legislation with three objectives. First, antitrust laws must be updated to allow AI firms to share information about evolving Chinese tactics to deter more threats. Second, the US needs more export controls on chips to hamstring Chinese access to advanced compute so that they simply can't train on US model outputs. That could make conducting distillation attacks pointless, Anthropic suggested. Finally, Congress should pass laws penalizing Chinese labs' "bad behavior" so that it's "more difficult and costly" to rely on distillation attacks to advance Chinese models. Penalties could include limiting Chinese firms from accessing US models or advanced US chips or from relying on data centers outside of China, Anthropic suggested.
[1] https://arstechnica.com/tech-policy/2026/06/anthropic-claims-alibaba-defied-trump-to-attack-claude-and-steal-capabilities/
[2] https://it.slashdot.org/story/26/04/07/2115208/anthropic-unveils-claude-mythos-powerful-ai-with-major-cyber-implications
[3] https://it.slashdot.org/story/26/04/22/2038241/anthropics-mythos-model-is-being-accessed-by-unauthorized-users
Good first attempt at protection... (Score:1)
Just curious....does Anthropic and other AI companies as a matter of policy prohibit connections from China or any of their co-horts?
Do they block all IP addresses from china?
Re: (Score:2)
Blocking traffic from China would do you no good. IP geocoding is only a guess at best. The 28,000 bot accounts will appear to be US-purchased smartphones, which will blend in with all the other traffic in every way possible. Detecting a botnet built by a company or government who knows what they're doing is very difficult. We used to be able to detect botnets, but they were mostly the DDOS kind, not the kind that steal data. Cloudflare is probably the best equipped to detect traffic patterns like this. If
Clown Shit (Score:1)
> Anthropic appears unconvinced, however, that Alibaba isn't working with the Chinese government.
No fucking shit. All the AI companies are working "with" their respective governments. So what?
Re: (Score:3)
Well, Anthropic isn't supposed to be. Anymore.
Unless you count asking your government to target your competitors of course.
"Working with the government" (Score:4, Insightful)
Yeah, they are. We know they are. That isn't magical knowledge. It happens in every country, and is mostly public knowledge. Guess who the US Military works with? Just look at publicly disclosed contracts. BAM, not that hard. Hell, AWS openly advertises its "Gov Cloud" region. Who the fuck do you think the "gov" is? Yeah. You think Alibaba doesn't do the same domestically? Of fucking course they do. Governments don't operate in vacuums, they have contracts with vendors to build shit, even if its the same shit a normal consumer can get.
Re: (Score:2)
China is a communist country. That's a whole other level of government involvement in the economy and specific companies such as Alibaba than you'll find anywhere in the US.
Re: (Score:2)
Every government has a way to compel it's corporations into compliance or death. Most are smart enough and patriot enough to choose to work with the government.
Re: (Score:2)
Your comment completely ignores the previous one. In a communist society, companies and the government are one entity. They are the same. That is fundamentally different from how companies and government works in capitalist societies.
Re: (Score:2)
Yes. In socialist societies, the government owns companies. In capitalist societies, companies own the government.
Anthropic needs to... (Score:2)
...stop spreading fear and calls for government control.
If government takes control, we all lose.
Re: Anthropic needs to... (Score:2)
It's your government, it exists because the people grant it the authority. Take control of it for once instead of acting like the government is a separate and opaque entity.
Re: (Score:2)
We voted. What would you have us do? Another January 6th? Guess it's okay if it's not your politicians making the questionable calls.
You're so cute when you think voting matters (Score:2)
I have a really funny idea for how to use computers to make voting better rather than ever more meaningless. Just as a tangential-topic typing exercise, I'll share it:
Still needs a good label, but the key ideas are ranked guest voting and equalized representation. Guest voting would allow you to pick from a neighboring district, which would be even more effective when your district has been aggressively gerrymandered. The electronic ballot would start with your own district, but you could click to get all t
Re: (Score:2)
The entire world is moving back toward aristocracy, but whereas China's aristocracy is based on being able to actually do stuff and Russia's is increasingly going to be based on having been successful at whatever you were doing in a war, the West's aristocracy is based on being able to convince the people with money to give it to you.
Re: (Score:2)
It's not Slashdot you need to convince. I'm quite sure if a committee of Slashdot's top 100 posters (we have that many?) could come up with better solutions then our rulers. Of course, most us likely want to improve things for humanity versus the rulers who are obviously working on maintaining their power and influence. It only looks like our rulers are idiots because they all lie about the agenda they are really for.
They (rulers) are also spectacular at pitting the masses against each other. It works very
Re: (Score:2)
> It works very well because the average person lacks critical thinking and would rather "fit in" then rock the boat.
So why would you let them vote?
And that's how China surpassed American chip perf (Score:1)
Anthropic suggests creating a massive demand for Chinese home-grown CPU and GPU development, and preventing American chip companies from selling and catering to it, in order to prevent China from catching up to US AI tech company level capabilities. That sounds like it will never backfire...
Re: (Score:1)
That's not going to make China catch up to "the US" because it's not China vs the US. It's China vs. the world. Literally no country wants China to be in charge of chip production except maybe Russia, and even then only because 1) they have absolutely no hope of being competitive themselves, ever , given the way they run their country; and 2) China is still happy to do business with Russia since no one can afford to sanction them for it.
Re: (Score:2)
China doesn't need to catch up to the US. They can simply invade and take over Taiwan where all the chips get made these days.
World's Biggest Raccoon (Score:3)
"Don't touch my garbage! I stole it fair and square!"
Is Claude benefiting from Chinese research? (Score:2)
Chinese research and model weights are published in the open and seem to be VALUABLE contributions to the field, according to 2min papers, (Deepseek not Qwen but just as an illustration): DeepSeek Just Solved AI's Billion Dollar Problem [1]https://www.youtube.com/watch?... [youtube.com]
[1] https://www.youtube.com/watch?v=mG4SmhWyeFA
Pot... meet kettle... (Score:5, Insightful)
All of these massive LLM companies were built on stealing copyrighted materials and other prior work to train their models (and then profit from it without paying for the data/images/etc), so the whole thing is hilariously thick with irony. "Hey, you can't steal the sh*t I stole!"
Now where did I put that teeny tiny violin...
Re: (Score:2)
Sounds like Apple and Microsoft fighting over Xerox technology.
You would think (Score:2)
This would be easy to stop, just find the offending accounts and inject garbage or nonsense into their streams. A human wouldn't care but an AI might if you were trying to train data on it.
Distillation is not an "attack" (Score:3)
"Second, the US needs more export controls on chips to hamstring Chinese access to advanced compute so that they simply canâ(TM)t train on US model outputs. That could make conducting distillation attacks pointless, Anthropic suggested."
Distillation requires relatively little compute. This is like trying to prevent the builder of a cruise ship from obtaining the means of obtaining deck chairs to prevent them from competing with other shipyards. This isn't how reality works.
"We believe combating the threat of illicit distillation requires coordinated action between government and industry, and we will continue working with Congress and the Administration to maintain American AI leadership"
"illicit distillation" is not a thing. In a way the crying is quite telling. If the statement is true and in fact "AI leadership" is based on distillation then Anthropic and everyone else is fucked.
If it was easy everyone would do it (Score:2)
If your product can be cloned, it will be. And the clones will be sold to your customers for less.
The only things you can do are:
Make the cost of reverse-engineering your products as high as possible. Don't make it easy. Include traps in your design to slow those who try to re-implement without fully understanding. If it is easy to copy your work, then they can sell clones of your work for less than you can afford to sell the original work.
Keep innovating. If the competition is a generation or more beh
Vizzini (Score:5, Insightful)
You're trying to kidnap what I've rightfully stolen!
Re: (Score:2)
By paying for it!
Re: (Score:3)
Not far from the truth.
The big implication from this is that making these inference engines isn't difficult and they can't be protected. That's the doom of all of AI money: the product is getting commoditized; you'll buy a Claude/ChatGPT/whatever in a box from Amazon in a few years.
Re: (Score:2)
They already have local models that run entirely on your own machine and those models are becoming more memory efficient, lowering the bar for the amount of hardware you need to get results. Given the spike in memory prices, I'm sure the developers will continue to strive for increase memory efficiency.
As you said, in the future, you'll get AI in a box from Amazon.
Re: (Score:2)
I think the end state of all this looks like game cartridges. If people could buy the weights of a frontier quality model in a high density, high speed ROM that ran locally plugged into an PCI-E or M.2 slot, all you would need then is a reasonably fast tensor processor with a little RAM for context. This is possible, and there is even a company (Taalas) with an early product. They have an online demo that is crazy fast.
You would buy one and use it for some time, probably years, and then discard it when
Re: (Score:2)
Of course. The point of machine learning is that the thing learning on its own is much cheaper than engineers figuring out how something works and hand programming a model. People have been talking about this since at least the 50s. It seems the mainstream needs to rediscover it every couple decades though.
It's not an attack, you silly human! (Score:2)
I checked with the AI and it's just research. Completely legitimate scientific work. The AI also said that Anthropic's lawyers must be smoking crack if they think the Chinese government is going to penalize or hinder any AI research in the Middle Kingdom.
Your mileage might vary depending on which AI you ask, but I actually suspect all of them agree on this one. But it is NOT any sort of conspiracy.
Why am I not surprised to hear that Microsoft is using cheap Chinese AI these days? Can't possibly be a lapse t
Re: (Score:2)
It's just preparing a narrative in the media. No different from what Anthropic did with Mythos and Fable. "Waaa, waa, we could so make money if the bad competitors didn't thwart us and bring our profits down by luring our customers with lower prices"
Remember: it's IPO season for Anthropic and OpenAI. Anything they say now is (more) suspect (than usual).