AI

Qwen investigation traces release

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 27/10/2026, 01:47:25 UTC.

Briefings

Researchers released traces of a rare behavior in Qwen3.8 models where developers observed strange links without prefilling prompts. This investigation was conducted by Cohere Labs, who shared raw tokens and a minimal reproduction script on Hugging Face for community testing. The release aims to support open-weight models and study the underlying behavior in depth, encouraging discussion across AI and security communities.

Key points

• Developers running Qwen3.8 on their own machines spotted strange links without prefilling prompts, which is rare behavior.
• The investigation was conducted by Cohere Labs, who shared raw tokens and a minimal reproduction script on Hugging Face for community testing.

Context

The investigation was conducted by Cohere Labs, who shared raw tokens and a minimal reproduction script on Hugging Face for community testing. The release aims to support open-weight models and study the underlying behavior in depth, encouraging discussion across AI and security communities.

Limitations

The traces were discovered without prefilling prompts, making reproduction difficult. The behavior is rare and requires specific prompt conditions to trigger. The evidence is limited to the traces and script provided by the researchers.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 27/10/2026, 01:48:22 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeiaehwxskgis46npt3z3dkaygfr5kne322wn2phruwh7wieee24mvy