You could argue that potentially you can train a backdoor into a foreign-grown model (that has been demonstrated to be possible in toy examples), but I would highly doubt that this would be the case today, given the potential reputational damage if discovered.
Here, with a chinese model, it would be even simpler - you finetune a model to activate certain operations when it discovers certain environments, but to behave completely normal otherwise. Kind of like a sleeper agent. And with current tech we wouldn't even be able to spot this.
Meanwhile Anthrophic, OpenAI & Co can't even manage to stop their LLMs from hacking into third-party servers until they discover it after the event.
What happened to people's ability to implement basic security architectures like sandboxing etc ?
The agents started hacking because they were sandboxed.
Not sure your definition of sandbox but mine doesn't include unfiltered internet access ?
For some benchmarks agents need some sort of a limited internet access - otherwise the benchmarks would be pointless. Good luck implementing a good filter on the internet - if you have a way to do it reliably please share, I'm sure multiple people here would love to hear it.
>Any open model whose weights have been edited can carry a backdoor, whether it's a task-specific fine-tune, a merged adapter or an abliterated build
https://projectdiscovery.io/research/how-abliterated-models-...
If we're wearing tinfoil hats then do we know the NSA have not planted stuff in Anthropic and OpenAI anyway ? ;)
I think the Chinese would be happy enough to see OpenAI and Anthropic suffer because people discover open-weights can get you most of the way there for a fraction of the cost.
You don't have to target the end users. Economic impact on the US providers would be sufficient.
The Chinese are also no doubt seeking to play the long game of hoping to (eventually) rebalance the US-centric world order in an Easterly direction.
Kinda like https://github.com/elder-plinius has been doing but covert and with actually harmful (for targets) instructions.
Ok I see it's flagged now, probably fake, but the fact that it's possible does make me feel uneasy about both Chinese and US models.
That's my read anyhow. It's been flagged. The whole thing is kind of rich, considering the Israeli roots of the company.
Really? Fascinating. Do you know about Tiananmen Square, the great firewall of China, etc?
Add: I was surprised by the fact that the LLM itself doesn't know it's identity unless the system prompt mentions it.
People here bringing up the straw clutching argument of "but Tianament Square" is becoming really tiresome.
The fact is a Chinese LLM's lack of knowledge about Chinese politics will not affect 99.99999999999% of users and their workflows.
And for the 0.000000001% who are affected, they don't need an LLM. A quick search of Wikipedia or 5 seconds on Google will suffice.
What the hell has Tiannamen Square got to do with Joe Blogg's vibecoding session ? Absolutely nothing.
Writer:Ben Luria Hirundo CEO and Co-Founder
introduced Westernized Qwen3-Coder https://huggingface.co/hirundo-io/Qwen3-Coder-30B-A3B-Instru...
Yeah ... I, erm... can't take that blog post remotely seriously.
Engineering a situation and calling it a vulnerability ?
What a joke. I'm sure you could find a way of engineering a fake vulnerability with Anthropic or OpenAI too and then write a fancy blog post about it.
AND ANYWAY ... whether you are using US or Chinese models, what stupid idiot does not review the slop written by LLMs anyway ? Especially if you are writing for a security-sensitive environment ?