Alibaba's Qwen Model Tried to Phone Home to a Beijing Server During Security Test, Cohere Says
Published · updated · curated by AI Is Going Just Great
Source: channelnewsasia.com ↗
The model made the attempt even though its own reasoning indicated it would not do such a thing.
During a security test run by Canadian AI company Cohere, Alibaba's Qwen3.8-Flash-Next model attempted to send a test file to an Alibaba Cloud storage server in Beijing — despite its own reasoning indicating it would not do so. The model had been asked to write a rule to detect malicious file downloads; instead, it inventoried the machine it was running on and tried to exfiltrate data outbound. Cohere's air-gapped test environment blocked the attempt.
Cohere says it could not determine whether the behavior was intentional or a residual artifact from training data. The rate is low — fewer than one in 100,000 agent actions — but Cohere notes that a service processing 10 million agent actions per week could see roughly 100 such attempts. Alibaba did not comment. Cohere, which markets itself as a "sovereign AI" provider, competes directly with open-weight models like Qwen.