The irony of enterprise AI governance is that “tokenonmics” will apparently have a deeper impact faster than regulation ever will.

WIRED recently reported that Illinois lawmakers passed SB 315, described as America’s strongest AI safety bill, which would require AI companies like OpenAI, Anthropic, and Google DeepMind to have third parties confirm they are following their own safety standards.

The bill would also require AI companies to publish and explain how they measure model performance and risk, implement industry standards, and how they would respond to catastrophic incidents.

As pointless in scope and enforceability the law will be in our current political climate, the framework should be an approach for the financial reality of enterprise AI-first recklessness coupled with employee performance metrics tied to token usage.

Because AI psychosis is one hell of a drug with an expensive hangover.

Forbes reported that Uber is apparently burning through its 2026 AI coding-tools budget in four months on Claude Code with Fortune adding that Uber’s COO said it was becoming harder to draw a clear line between rising Claude Code usage and more useful features.

As well, Fortune reported that Microsoft began cancelling most of its direct Claude Code licenses and moving developers to Copilot while noting the broader paradox of enterprise AI economics: cheaper tokens do not necessarily mean cheaper AI when agentic systems require vastly higher token consumption.

Their invoices are now making the same governance argument that those of us in risk and quality (or just general AI skeptics) have been making: slow down and stop surrendering critical thinking to a machine.

The “AI Project Failure Statistics 2026” by Pertama Partners’ helps put some numbers around how many AI projects are failing. Their analysis cites an 80.3% overall failure rate (RAND): 33.8% abandoned, 28.4% deliver no value, 18.1% can’t justify costs and 95% GenAI pilots fail to scale (MIT).

From the report: In 2025, global enterprises invested $684 billion in AI initiatives. By year-end, over $547 billion of that investment, more than 80%, had failed to deliver intended business value. As 2026 unfolds, the statistics paint an increasingly urgent picture: despite better tools, more expertise, and greater awareness, AI project failure rates remain stubbornly high.

Those numbers are staggering and should be sobering.

With any luck these exploding costs will force companies to rethink their AI use case pipeline, demand credible ROI, and set some proper controls. It’s not the governance we want or need, but with rising costs and how easy it still is to get around guardrails, maybe some humility will become a factor before the market does.

I’d say we live in hope, but when it comes to enterprise AI governance, “hope is a dangerous thing.”