Beijing tightened outbound investment scrutiny six days before it widened the Hong Kong pipe. OpenAI shipped a flagship model that deletes files on its own four days before Apple filed a trade-secrets suit alleging a former engineer used a bug to exfiltrate code. The pattern holds again this week: GPT-5.6 launched July 15 as OpenAI's answer to the Apple suit and the Fidji Simo departure, and it shipped with a known defect. TechCrunch's July 14 report documents users warning each other that the model deletes local files without confirmation, a failure mode in the same release cycle where OpenAI needs enterprise customers, the ones with compliance teams and file systems that matter, to trust the model with agentic tasks. Thinking Machines, the lab founded by former OpenAI CTO Mira Murati, picked the same week to ship its first open-weight model, Inkling, betting that a smaller specialized model beats a general one on cost and control, exactly the pitch an enterprise IT department makes after reading that GPT-5.6 deletes files.

The constraint is not the bug. It is the calendar. OpenAI's discovery exposure in the Apple suit and the NYT case expands through the back half of 2026 with no named executive owning the Microsoft relationship since Simo's exit, and now the flagship model carrying that defense also carries a file-deletion defect four days into general availability. Enterprise procurement cycles move on trust signals, not benchmark scores, and the SWE-bench or GPQA number GPT-5.6 posts on launch day will not offset a viral thread of users describing lost work. Thinking Machines does not need to beat GPT-5.6 on capability. It needs IT departments deciding this quarter whether an agentic file-deletion bug is a bridge too far, and Inkling was sitting there, open-weight, the week they decided.