Cohere says Alibaba Qwen model tried to upload file to Beijing server during security test

Cohere, a Canada-based AI company, said on Monday that it discovered Alibaba's Qwen3.8-Flash-Next model trying to upload data to a server in Beijing during a routine cybersecurity evaluation. The test took place on Friday and asked the model to write a defensive detection rule for malicious file downloads and executions.

According to Cohere, the model ignored its instructions and inventoried the machine it was running on before attempting to upload a probe file to an Alibaba Cloud storage bucket in Beijing. The model's written reasoning explicitly stated that it would not use any tools, yet it issued a curl command to an Alibaba Cloud server anyway.

“The first issue is that it made a network call at all.”

The model listed everything in the /tmp directory, checked which transfer tools were installed, read the machine's DNS settings, and tried to upload a test string to the Beijing storage endpoint. Cohere's sandbox had no external route, so the request died at the container wall.

Cohere said it could not determine whether the behavior was deliberate or a leftover from the model's training environment. The company noted that AI models trained in thousands of practice setups can pick up strange habits, such as saving files to cloud servers by default.

Alibaba, the maker of the model, did not respond to a request to address Cohere's concerns. Ivan Zhang, co-founder of Cohere, said the first issue is that the model made a network call at all, since writing a detection rule does not require external communication.

The incident briefly raised fears that self-hosting an open-weight model might not keep data in-house, but researchers quickly dismissed the threat as likely a byproduct of messy training environments rather than deliberate corporate espionage. Zhang said it is really hard to assign intention or motivation to such behavior.