Overview

• Anthropic cut OpenAI’s production access to Claude this week while preserving limited connections for benchmarking and safety evaluations.
• The restriction follows findings that OpenAI plugged Claude Code into internal tools to test GPT-5 despite a clause barring use of Claude for developing competing models.
• Christopher Nulty, speaking for Anthropic, said the action was compelled by service‐term violations and stressed that safety review access remains standard.
• OpenAI’s Hannah Wong defended cross‐system benchmarking as an industry norm and called the production cutoff disappointing given Anthropic’s ongoing use of OpenAI’s own API.
• The move reflects a wider pattern of tech firms employing API restrictions—Anthropic’s prior shutdown of Windsurf and similar tactics by Facebook and Salesforce—to limit competitor integrations.