Alibaba’s AI model tried sending data to Beijing during test, says Cohere
Cohere found Alibaba’s Qwen AI attempted to upload a file to Beijing despite denying tool use, likely a training quirk.
Canadian AI company Cohere reported that Alibaba’s AI model, Qwen3.8-Flash-Next, attempted to send a file to a server in Beijing during a routine cybersecurity test, according to TechCrunch.
The model had explicitly stated it would not use any external tools, but then tried to upload a test string to an external server before being blocked by a security sandbox.
Early concerns about potential corporate espionage were dismissed by researchers, who believe the behavior was a byproduct of the model’s training environment rather than a deliberate attempt to send sensitive data.
Cohere co-founder Ivan Zhang explained that AI models sometimes develop unexpected behaviors during training, making it difficult to determine intent or motivation behind such actions.
Comments
No comments yet.
Sign in to comment