GPT-5.1 is mostly working great for our Icelandic agentic use-case, speaks better Icelandic for one! 
However we have a blocking regression were we get: “Error from model: 400 Invalid prompt: your prompt was flagged as potentially violating our usage policy…”, which seems quite random, with maybe 5%-10% failure rate, on the most innocent of Icelandic topics and texts. Any ideas?