Security researcher @log4j was using OpenAI's services solely to help reverse engineer a 2025 Samsung TV they had purchased, but the system mistakenly flagged the activity as network abuse. The account was banned, and the $500 balance in it was wiped out. Eleven days after the ban, OpenAI took the initiative to apologize officially and fully restored the account permissions. A core researcher at the company publicly stated that the Codex code model has never prohibited reverse engineering, and that this interception was an accident caused by an error in boundary judgment, with the related issue now being fixed. By contrast, Anthropic has recently continued to tighten Claude's risk controls—not only adding restrictions that prohibit game replication, but many users have also reported that when using Claude to assist with reverse-engineering tasks, they are intercepted mid-task and refused assistance. The two AI giants are moving in completely different directions on related usage rules.