What Arts Orgs Should Know About Governing and Using Frontier AI Models

Q: How should arts organizations think about their reliance on frontier AI models vs. building their own tools?

A: Microsoft CEO Satya Nadella recently argued that every company should build its own AI model rather than relying on a handful of frontier models. His reasoning is that your organization’s knowledge is a competitive advantage, and you shouldn’t become dependent on a single AI provider or outsource your institutional learning.

I agree with the underlying principle. Your organization’s knowledge, processes, and expertise are valuable assets, and there is real value in building AI that understands your business rather than starting from scratch every time.

Where I struggle with is the practicality. Most organizations already trust third-party platforms with their most important information. Google has our email, Slack is where work conversations happen, HubSpot stores our CRM and marketing data, and Microsoft manages countless business documents. Building and maintaining a proprietary AI model simply isn’t realistic for most organizations, nor is it where I think they should invest.

I think the more practical takeaway is that organizations should focus on owning their AI layer, not necessarily their own model. Capture your institutional knowledge, build shared workflows, create reusable tools, and avoid becoming dependent on any single AI provider. For companies handling highly sensitive or regulated data, building a private model may absolutely make sense. For everyone else, the competitive advantage is less about owning the model itself and more about owning the knowledge and processes that sit on top of it.

Q: Have any new voices been added to the larger AI governance conversation?

A: Google DeepMind CEO Demis Hassabis is one of the latest AI leaders to publicly discuss the magnitude of the impact these models could have on the world and the need for stronger governance. His proposal calls for creating an independent body to evaluate the most capable AI models before they’re released.

For me, his specific proposal is less important than the broader signal. Over the past few months, we’ve heard similar themes from leaders at OpenAI, Anthropic, and now Google DeepMind. While they may disagree on the details, they’re increasingly aligned on one thing: as AI capabilities continue to advance, we need better frameworks to ensure these models are developed and deployed responsibly.

Whether or not Hassabis’s proposed solution is the right one, I think it’s notable that the people building the most advanced AI systems are consistently saying that governance needs to evolve alongside the technology.

More to come,
Jen

Have a question for a future edition? Submit it here!