OpenAI puts a 230-million-URL legal index behind Astra for Law

If you build legal tooling, the decision is whether to design against the plugin surface now or wait for the API. OpenAI says API access comes later, so anything you wire today runs inside ChatGPT or Codex under Trusted Access and cannot be assumed to survive as-is when the endpoint ships. Treat the index as retrieval, not authority: keep a citation check that resolves every quoted authority back to a primary source and records which plugin returned it, because partner and community plugins are different trust boundaries. If you already run legal retrieval, hold your current evaluation set constant and run it against both paths before moving anything.

The eligibility question comes before the capability question. Trusted Access is invitation-based, so for most firms the practical answer this quarter is that the product is unavailable, and there is no published price to compare against an existing research subscription. Two further limits are stated: the index is US law, and the announcement does not address Europe. If you handle client matter data, get the retention and human-review terms in writing rather than inferring them from a launch post, and confirm with your risk committee that professional-responsibility duties on verifying authority sit with the lawyer regardless of what the tool cites.