Beyond Shadow AI: What Is on the AI Governance Leader’s Desk Now

By Dóra Cseresnyés, Managing Partner, Hybridge AI Governance. Originally published on LinkedIn on 26 October 2026.

In many organizations, AI governance conversations still start with the same problem. Employees use AI tools nobody has approved, and governance teams spend much of their energy proving that they need to exist at all.

Some companies, mostly larger ones, have moved past this stage. Their governance function is up and running, and shadow AI is largely under control. For them, the harder work is making the framework run.

EY’s latest AI Risk and Governance Survey captures this stage well: companies are formalizing governance while rolling out AI and autonomous agents at speed, and many still struggle to keep policy and practice aligned.

So what does an AI governance leader work on once that foundation is in place? Looking at my own agenda and at the conversations I have with peers, five topics stand out. Each of them is a business question as much as a compliance question, and each one shapes decisions that used to sit far away from governance.

1. Preparing for the EU AI Act on a product timeline

The Digital Omnibus, in force since July 27, 2026, moved the deadline for the high-risk AI systems listed in Annex III to December 2, 2027, and for AI embedded in regulated products to August 2, 2028. Many leaders read this as breathing room. I read it as a planning date. A product that enters design today will typically reach the market in 2027 or 2028, exactly when these rules start to apply, and compliance designed in from the start is far cheaper than compliance retro-fitted after launch.

This is why the preparation rhythm should follow the product roadmap rather than the regulatory calendar. Risk classification belongs at the concept stage, because it decides which obligations the product will carry for its whole life, and documentation can then grow alongside development.

2. EN 18286 and ISO 42001: knowing which one you need

On October 14, 2026, EN 18286 became the first harmonised standard under the AI Act to be referenced in the Official Journal of the EU. It turns the quality management system requirements of Article 17 into auditable requirements for providers of high-risk AI systems, and applying it gives a presumption of conformity with those provisions. ISO 42001 plays a different role. It is a certifiable international standard for an AI management system, and it carries no presumption of conformity under the Act.

The distinction leaders need is organisational governance versus system-specific conformity. ISO 42001 builds the first: a certifiable management system that shows partners the organization governs AI in a structured way. EN 18286 serves the second, and for providers of high-risk systems it is the route to demonstrating legal conformity of the quality management system. The two standards were designed to fit together, so an existing ISO 42001 system is a foundation to build on.

EN 18286 is also only the first of several. Harmonised standards on topics such as risk management, data governance, bias, logging and human oversight are still in development, and each will bring new work once it is published. This work is best spread out continuously, building on the existing management system as each standard arrives, rather than handled as a single project right before a deadline.

3. Vendor vetting: you can buy the system, not the accountability

Most companies use AI built by someone else, which makes them deployers under the AI Act, and deployer obligations cannot be outsourced. For high-risk systems, the deployer has to use the system according to the provider’s instructions, assign trained people to human oversight, keep logs for at least six months and inform employees before the system goes live. None of this is possible if the vendor contract does not secure access to documentation and logs. Incident handling belongs in the contract as well, since a deployer who identifies a serious incident has to inform the provider and the authorities without delay.

There is one more trap worth flagging to the executive team. A company that puts a third party AI system on the market under its own name or trademark can become a provider in its own right, with the full set of provider obligations. Vendor vetting is therefore a business risk process that runs for the whole life of the contract. A structured third party vetting process, with clear criteria before signing and regular reviews after, is one of the most valuable things a governance team can build.

4. Goal alignment in the age of agents

Agents have arrived faster than the frameworks meant to govern them. In the EY survey, 91% of senior AI executives say their organization uses agentic AI, yet about half of those report that their governance framework has not been updated for agent specific risks. A quarter cannot even detect unauthorized agents running inside the organization. Shadow AI has followed the technology and moved to the agent level.

This calls for a new division of labour. IT owns the technical controls, such as identity and access management for every agent. Governance controls matter at least as much: what each agent is trying to achieve, and whether that goal matches what the organization actually wants. An agent optimizing a narrow target, such as faster ticket closure or lower cost per claim, can drift away from customer outcomes without breaking a single technical rule. Defining counter-objectives and forbidden paths is just as important as defining the goal itself. Governance also has to decide which decisions an agent may take alone and which need human approval, and every agent needs a named business owner who answers for its goals.

5. Board reporting: governance as decision support

Everything above ends up in one place: what the board and the executive team actually see. PwC’s 2026 Annual Corporate Directors Survey found that 82% of directors consider the information linking AI outcomes and risks to business performance inadequate or missing. That gap belongs to governance. A useful board report connects each material AI system to an accountable owner and a measurable business outcome, and it shows how the risk picture has changed since the last meeting.

Good reporting turns governance into an early warning system. A board that sees where AI creates value and where it creates exposure can back the right initiatives with confidence and pull back from the wrong ones before they turn into costly incidents or lost customers. This is how governance reduces business risk in practice, and it protects leaders personally as well, since oversight is judged by what they knew and how they acted. At that point governance is no longer a reporting function. It is part of how the company decides.

The common thread

Across these five topics, AI governance plays two roles at once. It steers the organization’s use of AI toward compliance, and it supports leadership decisions: when to launch a product, which vendor to trust, how much autonomy to give an agent and what the board needs to see. That is how governance earns its place at the leadership table, as a partner in the decisions that shape the business.

Sources

  • EY survey finds that autonomous AI implementation outpaces oversight, yielding an AI governance gap (EY, September 2026)
  • First ever European standard on AI published in the OJEU (CEN-CENELEC, October 2026)
  • 2026 Annual Corporate Directors Survey (PwC, September 2026)

Governance that starts small and grows with your organisation. See where you stand in three minutes.

Take the AI Governance Quick Scan →Explore FairPath →