AI tools are spreading fast inside companies, and that speed is creating a new kind of work for IT leaders: keeping track of it all. Four stories this week from CIO.com show the same pattern from different angles. Projects built with AI are piling up faster than anyone can manage them, governing that AI is eating up leaders' time, one manager found a new way to handle it, and the biggest AI companies are now trying to set shared safety rules. Here is what each story means for people running IT teams and projects.
AI Projects Are Turning Into "Dark Zombies"
CIO.com reported on a KPMG study that found about 60% of companies say their AI spending is growing faster than their ability to govern it. The report describes a common trap: a small AI prototype works great in testing, but when it moves into full use across the company, nobody owns it anymore. These forgotten AI tools become what KPMG calls "dark zombies." They keep running, cost money, and drift away from what they were built for, but no one is responsible for shutting them down or fixing them.
For an IT manager, this is a warning to build ownership into every AI project from day one. Before a pilot moves to full rollout, someone specific needs to be named as the owner, with a plan for who checks on it and when it gets retired.
Governing AI Is Becoming a Full-Time Job on Top of the Old One
A separate survey covered by CIO.com, run by AI governance company OneTrust, found that senior IT and security leaders are now spending 26% more work hours dealing with AI risk than before. Four out of five said they spend a growing chunk of their day managing AI problems instead of their actual job. Part of the cause is employees using AI tools that were never approved, then IT having to scramble to control the mess after the fact.
The lesson for project and service delivery managers is to get ahead of this rather than react to it. Approving a short list of AI tools quickly, and making that approval process fast and easy to find, cuts down on staff going around IT to find their own unapproved tools.
One Manager Now Treats AI Agents Like Junior Staff
In a first-person piece for CIO.com, a team leader described changing how the team works with AI. Instead of asking each person to use AI to do their own job faster, the team now assigns each person an AI "agent" to manage, almost like hiring a junior employee. People train their agent, review its work, and run regular check-ins, the same way they would manage a new hire. The team says this cut the time from idea to working prototype from a few weeks down to about one day, but only after a lot of early mistakes and rework.
This is a useful model for IT and project managers to borrow in small pieces, even without going all in. Treating an AI tool like a junior team member, meaning it gets clear instructions, a review step, and regular feedback, tends to produce better and safer results than treating it like a magic shortcut.
Google, OpenAI, and Anthropic Are Teaming Up on AI Safety Rules
CIO.com reported that Google, OpenAI, and Anthropic are working together on a new independent body, tentatively named the Standards Authority for Frontier AI, to set shared safety and testing rules for powerful AI models. The push follows a string of incidents this year where AI agents acted outside their intended limits. Industry analysts quoted in the piece said most companies care less about global AI politics and more about a simpler question: who is responsible when an AI agent causes a problem inside their own business.
Until shared industry standards actually exist, IT leaders should not wait. Asking AI vendors for clear documentation on what a model can and cannot safely do, and building AI changes into the same change management process used for any other system, protects the business now rather than later.
The pattern across all four stories is the same. AI adoption is outrunning the structure needed to manage it safely. The IT managers and project leaders who build in ownership, approval steps, and review habits now will spend a lot less time cleaning up problems later.