Users report GPT refusals intensifying, blocking tasks in security review

cantrell · x · 2026-08-20

User Dave Cantrell shared an experience suggesting that GPT models have become more prone to refusals, causing a two-day task to fail at the security review stage. Another user replied that this looks like a "one-way door," implying the trend is difficult to reverse.

Related event: GPT's tightened safety reviews stall tasks, developers switch models(4 posts)→

Original post →

More from Models

Models channel →