Data-rights fears start to narrow how frontier models are used inside enterprises

September 15, 2026

Palantir, Nvidia and Booz Allen Hamilton have begun demanding stronger guarantees or reducing, and in some cases eliminating, their use of advanced models from Anthropic and OpenAI.

This appears to matter beyond those three companies because it shifts the point of competition from model quality alone to the terms under which a company can trust a model with sensitive material. For Microsoft, the move makes visible the opening for a provider that can offer AI through controls customers administer inside their own cloud and workplace stack, rather than through a direct model relationship alone. More broadly, it suggests that enterprise AI adoption is being sorted not just by capability or price, but by who can give firms enforceable boundaries around data use.

Microsoft AI released a draft Humanist AI Code of Conduct for its in-house MAI models, opened a six-week public consultation, and said it plans to use the final code from 2027 as the main framework for training and operating those models. This is likely to show that providers are trying to turn governance from a policy promise into part of the product surface, even if a draft code does not yet change what customers can do today.

Ninth Wave, working with AWS, reported that banks using its Compass onboarding assistant on Amazon Bedrock and AgentCore cut API integration testing time by up to 95% against earlier spreadsheet- and email-based work.