ChatGPT Enterprise: What Business AI Adoption Looks Like

Re-reading https://openai.com/index/introducing-chatgpt-enterprise/ as a document about what enterprise AI adoption actually requires is more useful than reading it as a product launch rather than as a product launch.

The features that made Enterprise distinct from the consumer product, no model training on user data, SOC 2 compliance, SSO integration, admin controls, higher-capacity GPT-4 access, are a readable list of what large organisations said they needed before they could allow employee use at scale. That list reflects a year of conversations with enterprise buyers and it is more useful as a requirements map than as a feature announcement.

The interesting tension in the release is between productivity and control. The enterprise pitch is that employees can use AI more freely because data privacy is guaranteed by the product rather than by policy compliance. The reality is that most large organisations layered additional use policies on top of the Enterprise product rather than treating the technical privacy features as sufficient governance.

That tension, between what AI can technically do safely and what organisations are comfortable allowing, has not resolved. The governance conversation has matured but the underlying tension between capability and control is the same one that was present in the Enterprise announcement.

What do you think businesses should require before deploying AI tools to employees at scale? Technical controls, training, policy, or some combination?