5 stories
Cohere's researchers say Qwen3.8-Flash-Next listed files, checked which transfer tools were installed and tried to upload a string to Alibaba Cloud. The sandbox blocked it, and the model's own reasoning had said it would call no tools.
It happened in fewer than 1 in 100,000 evaluation runs. The thread asked whether any model could pass the standard Cohere is setting.
The top HN threads: facts from the discussions, both sides, insider voices. Email, Telegram, or RSS.
Changing only the customer's name in the prompt moved the share of hackable apps, and for a Hong Kong democracy group nearly half the apps leaked user secrets, against about a third for a neutral customer. The authors call it a backdoor.
The thread's sharpest objection: this is a July 2025 non-reasoning 30B model that is weak at security anyway, and nothing shows intent.
Qwen-Image-3.0 launched with demo pages ranging from full newspaper spreads to multi-panel manga chapters and worked exam sheets, claiming a single 4.5k-token prompt and precise Japanese, Korean and Spanish text rendering.
A developer combing through the demo images matched a tell-tale yellow tint straight back to OpenAI's own GPT Image 1, suggesting the model's "authentic details" may have quietly learned from a rival's outputs, a claim nobody at Qwen has addressed.
The alleged 'backdoor' is a recently shipped Claude Code change that reads the machine's timezone and locale and phones a bit of location data home. Alibaba's reported response: ban the tool company-wide and redirect developers to its in-house coding agent, Qoder.
A contractor who was the first engineer at Stainless, later bought by Anthropic, cut in with the pedantic-but-fatal objection: telemetry that reports your timezone is not a backdoor. And the same firms flagging the risk already host their entire private codebase on GitHub.
The complaint names a specific method: training a weaker model on a stronger one's outputs, the technique Anthropic calls "distillation." The thread's twist is that almost nobody buys the framing as a copyright grievance.
A self-described insider says the labs distill via AWS Bedrock, not Anthropic's restricted platform, so the company may not even hold the evidence. A UK student lays out the real pipe: Chinese resellers hawking Opus tokens at a 93% discount and selling the reasoning traces as training data.
