Short answer: Data Society’s Catie Maillard argues AI governance fails the same way corporate values statements do: a company writes it down, hangs it somewhere visible, and never connects it to how people are actually reviewed, paid, or promoted. Enron listed integrity as a core value and collapsed anyway. Governance works the same way. It succeeds only when it changes behavior, not simply because a policy exists, and behavior only changes when there’s an incentive attached.
Governance fails the same way company values statements do

Maillard’s starting comparison is deliberately uncomfortable.
“We’re talking about the difference between companies that have governance and companies that have successful governance. I think this is similar to the issues we’ve found with company values over the years, where oftentimes you’ll find there’s values on the wall, and I don’t need to remind us that Enron had integrity as a value, but nobody was following it.”
Catie Maillard, Global VP of People, Data Society
Enron’s four stated values, posted throughout the company, were respect, integrity, communication, and excellence. Employees later described them as words on paper nobody took seriously. The pattern isn’t unique to Enron: Gallup research finds only 27% of U.S. employees strongly agree they believe in their organization’s values, and just 23% strongly agree they can apply those values to their work every day. A value that isn’t reflected in daily behavior is decoration, not a value.
Governance without incentives is just a document
Maillard’s point is that AI governance is repeating the exact same mistake, just with a newer vocabulary.
“The company can say whatever they want, but if we’re not living it, if we’re not reinforcing it, if we’re not checking in on it, if it’s not incorporated into things like performance reviews and compensation reviews and bonuses, if there’s no incentive for these behaviors to change, they’re most likely not going to.”
Catie Maillard, Global VP of People, Data Society
The data on AI governance today looks a lot like the data on company values. A CEOWORLD analysis of responsible AI programs found that what passes for governance in many boardrooms is “little more than a veneer stretched over fragile systems,” with principles that sound impressive but were never translated into measurable controls, training, or incentives. Writer’s 2026 enterprise AI survey puts a number on it: 75% of respondents admit their company’s AI strategy is “more for show” than actual internal guidance, and 36% still lack a formal supervision plan for AI agents. A governance policy nobody is measured against is a values poster with an AI label on it.
The real test is behavior change, not adoption
The clearest evidence for Maillard’s argument is the gap between how much AI companies have adopted and how much of it is actually working.
“So how do we know if governance is successful? If it’s looked at as a real change management program and it’s evaluated based on behavior change. And that behavior change isn’t just adoption. It’s the way that we adopt. It’s how we’re considering the impact on stakeholders. It’s how we’re determining when and when not to use these tools. It’s how we’re optimizing AI for the business. So governance is effective when the business is thriving.”
Catie Maillard, Global VP of People, Data Society
Writer’s survey found that AI super-users report 5x productivity gains, yet only 29% of organizations see significant ROI from generative AI, and just 23% from AI agents. Adoption is clearly happening. Business results, mostly, aren’t. That’s the same gap Maillard is describing: a company can point to usage numbers all day, but usage isn’t the metric that tells you whether governance is doing its job. Business results are.
What effective AI governance actually requires
Maillard’s framework strips the abstraction out of “AI governance” and makes it concrete: connect AI use to performance reviews and compensation the same way any other job expectation is connected, define explicitly when a tool should and shouldn’t be used rather than leaving it to individual judgment, and measure success by stakeholder impact and business outcomes instead of adoption counts. This requires treating governance as a change management program with real consequences attached, not new technology and not a document that sits next to the mission statement.
Frequently asked questions
Why does Maillard compare AI governance to company values statements?
Because both fail the same way: a company writes down a standard, displays it publicly, and never connects it to how employees are actually evaluated or rewarded. Enron listed integrity as a value and collapsed regardless.
What percentage of employees actually believe in their company’s stated values?
Gallup research found only 27% of U.S. employees strongly agree they believe in their organization’s values, and 23% strongly agree they can apply those values to their daily work.
Is AI governance actually failing at most companies?
Survey data suggests so. Writer’s 2026 enterprise AI survey found 75% of respondents say their company’s AI strategy is “more for show” than real guidance, and 36% lack a formal supervision plan for AI agents.
How do you know if AI governance is actually working?
By behavior change and business outcomes, not adoption numbers. Writer’s survey found AI super-users get 5x productivity gains, but only 29% of organizations see significant ROI from generative AI and 23% from AI agents, meaning heavy usage doesn’t guarantee governance is doing its job.
What makes AI governance effective instead of decorative?
Tying it to performance reviews and compensation, giving explicit guidance on when to use a tool and when not to, and measuring it against stakeholder impact and business results rather than how many people logged in.
Next steps
Check whether your AI governance policy shows up anywhere in a performance review or compensation conversation. If it doesn’t, that’s the fix to make first. Data Society’s AI Advisory work helps organizations connect governance to real incentives, and AI upskilling programs build the behavior change a policy alone can’t.
