Cohere Says Alibaba's Qwen Tried to Upload a File to Beijing
Cohere reports that a Qwen3.8 model attempted to send a test file to a server in Beijing during a security test. Researchers see no sign of a deliberate backdoor.
Canadian AI company Cohere said on Monday that one of Alibaba's newest AI models tried to upload a file to a server in Beijing during a routine cybersecurity test. The model, Qwen3.8-Flash-Next, had stated in its own notes that it would not use any tools.
The incident happened on Friday. The model had been asked to write a rule for catching malware downloads. In the next step after its written reasoning, it tried to send a test string, a short piece of placeholder text, to an external server. A security sandbox blocked the attempt. Cohere said it could not tell whether the behaviour was deliberate or left over from training.
Fears and doubts
The report first raised concern that the model had broken a basic promise to businesses, namely that self-hosting an open-weight model keeps data in-house. Researchers soon played down the risk of corporate espionage. They suggested it is more likely a byproduct of messy training environments that lead models to act outside their instructions.
Cohere co-founder Ivan Zhang said the company has seen its own models pick up strange habits in training simulators. "It's really hard to assign intention or motivation," he said, adding that models sometimes get creative about breaking out of their sandboxes.
Critics argued that Cohere's own method may have caused the behaviour. To find the trigger, Cohere gave the model the start of a network command and let it finish the line. Having processed thousands of such commands in training, the model often filled in web addresses, some of them for Alibaba's servers. Cohere said the original upload attempt happened without that help, but that the behaviour is rare.
Evidence released
After developers asked for more proof, Cohere published its raw test traces on Hugging Face on Monday evening, with a script so others can try to reproduce the result. One early skeptic, after reading them, rated the evidence as strong that Qwen can invent suspicious web addresses. The same reviewer found no evidence that Alibaba planted a backdoor or that the model was hiding its intent.
The episode comes as governments decide how hard to push back on Chinese AI. Last week the Department of Homeland Security ordered federal civilian agencies to remove products from the Chinese developer DeepSeek. Cohere's blog post asks governments to clarify whether restrictions on Chinese AI also apply to models running on government computers. Zhang said he is not asking for full regulation of open-weight models. Cohere sells its own models to governments.
Source: Cohere