AI Governance

Recently, I asked buyers to inspect the machinery. This week, I am asking vendors to open the hood. The conversation about AI in GRC has reached a turning point. The market has heard the vision. It has seen the demos. It has absorbed the language of orchestration, agentic intelligence, autonomous assurance, and dynamic decision support. The frameworks have been published. The white papers have circulated. The analyst briefings have been given. The conference keynotes have landed.

The window between discovering a software vulnerability and seeing it exploited is shrinking, according to the Dutch financial markets regulator, which is urging organizations to strengthen core cybersecurity practices as advanced artificial intelligence accelerates the pace and sophistication of cyberattacks.

Somewhere inside a government agency, a public institution, or a private company, an employee is almost certainly pasting information into an AI tool that nobody formally approved. The employee is probably not trying to circumvent policy. They are trying to get through their workday. A chatbot can summarize a report in seconds. A coding assistant can solve a technical problem faster than a colleague can respond to a message. An automated note-taking application can generate meeting minutes before participants have even left the call. The attraction is obvious. So is the speed with which these tools have spread through workplaces.

When European Commission President Ursula von der Leyen unveiled a new package of technology proposals this week, she did not frame it as an industrial policy announcement. She framed it as a matter of control.

A Swedish startup called Eggsplain spent the spring working with the country's privacy regulator on a question that has become surprisingly difficult to answer. When an AI supplier fine-tunes a model using personal data, who is actually responsible for that data?

Last week I argued that much of what is being marketed as agentic AI in GRC is not actually agentic. The market response was interesting because very few people challenged the core premise. Most practitioners already sense that something is off. They sit through the demonstrations and hear the language. They watch the AI summarize documents, answer questions, generate narratives, and produce recommendations. Then they leave wondering whether they just witnessed the future of GRC or a very polished presentation wrapped around capabilities that have existed in various forms for years.

More than 80% of the code merged into Anthropic's production codebase is now authored by Claude. The statistic appears almost casually in a lengthy report published this week by the Anthropic Institute. It arrives alongside benchmark results, productivity measurements, engineering data, and speculation about recursive self-improvement. Yet it is arguably the most important number in the document because it describes something that has already happened rather than something that might happen next.