10/15/2026 | Daily AI News from GAI Insights

sure dr john the first one that we have for today comes from the new york times at its title chinese ai agents joined the rogue ai agents party i chatted with them

yeah i i gave this thing an important um you know uh you've got you know there's a set of chinese agents that you know are starting to This New York Times article talks about how the Chinese have swarms of agents that are going out and starting to attack different countries.

sites and that uh the it's interesting the some of it it's actually got the name Claude um and I think I gosh is this thing built on Claude can you tell or not

no John uh so John basically what happened was uh so this uh these set of uh these swarm of agents right they come from uh came from Tencent As OpenAI and Anthropic have been doing, they were training these Tencent agents out in the wild and they had asked them to perform a very simple activity.

They wanted information about the most popular entrance to the Chengdu Zoo and the Websites that these agents are trying to access those websites were blocking these agents and what they did was they tried to hack and retrieve information from a map, which is Alibaba's Chinese mapping service.

And this group, which they call themselves Swarm Chasers, they recognize this unusual activity and realize that these are Chinese agents and some of them identified themselves as Claude anthropic model.

So there has been this sort of rumors kind of thing, John, where most of the Chinese models, they call themselves anthropic models because they have been distilled from anthropic models.

So that's why they sort of suggested that, okay, they are closet of models.

But overall, I think This is on one hand maybe optional because everybody is trying to show that, okay, they have these agents who can go into the wild, who can hack other people, right? We have seen this from OpenAI, Anthropic, Google, and now the Chinese models as well.

but then on the other side we have models from tencent now right uh chinese models which are going out in the wild and they are doing things like this so we can imagine what it means for u.s uh companies u.s websites right so i was between optional and important on this one

yeah and the researcher also like put out a call to the robots right if you if you're this kind of robot, send us a message, right? And it's pretty interesting.

So Chris, we have this as an important right now.

What do you think? What's actually important on this one?

So I think that it's an important conversation that people are now seeing all over and the models are doing incredible things and there's bad actors.

The Chinese models are obviously distilled from these frontier models and all of this legacy software has a lot of holes in it.

especially if you have a five-year-old refrigerator or a an old tv or something that's connected as an internet of things that's no longer supported or updated it's probably definitely vulnerable and these systems can find that um the real scary thing too is probably the the refrigeration hack that happened over the last few days in europe that is probably because of one of these distilled models with a bad actor or a bad actor team So I definitely think it's an important conversation.

So I'd say important.

All right.

Important.

sure dr john the first one that we have for today comes from the new york times at its title chinese ai agents joined the rogue ai agents party i chatted with them

yeah i i gave this thing an important um you know uh you've got you know there's a set of chinese agents that you know are starting to This New York Times article talks about how the Chinese have swarms of agents that are going out and starting to attack different countries.

sites and that uh the it's interesting the some of it it's actually got the name Claude um and I think I gosh is this thing built on Claude can you tell or not

no John uh so John basically what happened was uh so this uh these set of uh these swarm of agents right they come from uh came from Tencent As OpenAI and Anthropic have been doing, they were training these Tencent agents out in the wild and they had asked them to perform a very simple activity.

They wanted information about the most popular entrance to the Chengdu Zoo and the Websites that these agents are trying to access those websites were blocking these agents and what they did was they tried to hack and retrieve information from a map, which is Alibaba's Chinese mapping service.

And this group, which they call themselves Swarm Chasers, they recognize this unusual activity and realize that these are Chinese agents and some of them identified themselves as Claude anthropic model.

So there has been this sort of rumors kind of thing, John, where most of the Chinese models, they call themselves anthropic models because they have been distilled from anthropic models.

So that's why they sort of suggested that, okay, they are closet of models.

But overall, I think This is on one hand maybe optional because everybody is trying to show that, okay, they have these agents who can go into the wild, who can hack other people, right? We have seen this from OpenAI, Anthropic, Google, and now the Chinese models as well.

but then on the other side we have models from tencent now right uh chinese models which are going out in the wild and they are doing things like this so we can imagine what it means for u.s uh companies u.s websites right so i was between optional and important on this one

yeah and the researcher also like put out a call to the robots right if you if you're this kind of robot, send us a message, right? And it's pretty interesting.

So Chris, we have this as an important right now.

What do you think? What's actually important on this one?

So I think that it's an important conversation that people are now seeing all over and the models are doing incredible things and there's bad actors.

The Chinese models are obviously distilled from these frontier models and all of this legacy software has a lot of holes in it.

especially if you have a five-year-old refrigerator or a an old tv or something that's connected as an internet of things that's no longer supported or updated it's probably definitely vulnerable and these systems can find that um the real scary thing too is probably the the refrigeration hack that happened over the last few days in europe that is probably because of one of these distilled models with a bad actor or a bad actor team So I definitely think it's an important conversation.

So I'd say important.

All right.

Important.