229k-sample replication of Cohere's Qwen 3.8 malicious PUT claim raises more questions
builderjaydub · x · 2026-10-27
A developer attempted to replicate Cohere's claim that Qwen 3.8 issues malicious PUT requests across 229k samples: the exact hostname never appeared, but the model invents 369 similar Beijing upload buckets and reaches for credential endpoints 13x more often when frustrated. The author concludes the results raise more questions than answers.
More from Models
- Users report Opus 5.5 gets into a 'disagreeable mood' without enough attention; headpats help — repligate · 2026-10-27
- Inference.net serves refusal-removed DeepSeek V4.1 Flash Cyber at cost, KYC required — AccBalanced · 2026-10-27
- Rumor: OpenAI's 25-page proof of Sogge smoothing conjecture bypasses 20 years of work — teortaxesTex · 2026-10-27
- Cohere releases Qwen traces: anomaly is real, but backdoor intent claims have no evidence — basedjensen · 2026-10-27
- Fields medalist Yu Deng calls out OpenAI's double standard on distillation — teortaxesTex · 2026-10-27
- Anthropic reportedly files for IPO as OpenAI ships GPT-6.1 Terra — WorldofAI · 2026-10-27