Anthropic discovered OpenAI’s engineers were using Claude Code to benchmark GPT-5, violating its commercial terms.
