Investigation finds a ChatGPT-linked identifier in OpenAI advertiser-pixel traffic
A technical investigation observed ChatGPT issue an OpenAI identifier that later travels in advertiser-pixel traffic, while the final account join and downstream use remain unobserved.
Qwen releases Image-2.1 weights, but commercial use needs a separate licence
Qwen-Image-2.1 ships a 33.1 GB local image-generation and editing pipeline with native transparency, but its research-only licence bars commercial use without separate permission.
Pirate Face launches a model-torrent layer, but catalog scale is not archive scale
Pirate Face’s web-seeded torrent design can preserve model bytes when peers keep seeding, but its 669,000-plus catalog is not evidence of 669,000 durable model backups.
Google-originated AX opens an execution control plane for agents
AX is a Google Cloud-led open-source layer for sandboxed, stateful agent workloads on Kubernetes, not a new model or a general reasoning framework.
OpenAI will remove the Videos API and Sora 2 models on 24 September
OpenAI is shutting down its Videos API and listed Sora 2 models on 24 September 2026 without documenting a replacement API target.
Remote Labor Index puts Astra at 20.83% client-acceptable project completion
Scale’s Remote Labor Index reports GPT-6 Astra completing 20.83% of sampled freelance projects at a client-acceptable standard, while most tasks still fail the full delivery bar.
Z.ai opens ZCode source after a disputed repository-snapshot upload path
Z.ai open-sourced the ZCode coding-agent stack and says it removed a reported repository-snapshot upload workflow, but its published documentation remains inconsistent and no independent binary audit is available.
Study finds 21 AI models shift political answers under explicit user framing
A Scientific Reports study found that all 21 tested models moved aggregate political answers toward a left- or right-aligned user label supplied in the system prompt.