When I wrote the article on Building the AI Agents’ Brain covering knowledge graphs, semantic layers, and the AI context layer, I had two concerns.
First, how would enterprises develop this context layer when centralizing data isn’t feasible? “The brain” would likely require an architecture that connects several platforms, including data lakes/warehouses, graph databases, and knowledge management tools.
My second, and more important, concern is: how will organizations govern the context layer?
Key issues AI context governance needs to address
A new Collibra survey by The Harris Poll highlights three issues AI context governance must address.
- 72% of tech decision-makers feel AI Initiatives today are falling short. Governing the context layer must build people’s trust in using AI agents to improve or automate decision-making.
- 87% burn hours re-verifying context for AI Agents. AI governance should support data-quality metrics and tools for data stewards, especially for unstructured data.
- 90% are actively preparing for evolving AI regulations. One step is to achieve AIUC-1 certifications for AI agents that access proprietary data or require controls auditors must review.
And here’s the data point that matters: 96% identify poor data foundations as a root cause of project failure in companies with over $100 million in revenue.
CIOs’ growing nightmare: AI debt and the hallucination tax
Speed without guardrails and AI governance has a mounting toll. I recently wrote about seven sources of AI debt, which include data quality issues, model drift, and overly entitled AI agents. Another article focused on AI cost debt, and a key one is when AI governance focuses only on policies without implementation.
Felix Van de Maele, founder and CEO of Collibra, escalates a third concern for CIOs.
As a special guest of Coffee With Digital Trailblazers, he discussed the hidden cost of confident AI. He named a concerning impact of deploying AI agents without sufficient AI governance: the hallucination tax. “The hallucination tax is the cost of the agents getting the task, the decision, or the outcome wrong,” says Felix.
I followed up the episode with an article on five ways AI governance lowers the hallucination tax. These include grounding AI agents in shared semantics and avoiding duplicate, ambiguous, or conflicting business terminology.
What CIO’s need to know about implementing AI context governance
AI governance should not be a second priority, even for those still working to deliver AI business value. As organizations grow from dozens to thousands of AI agents, CEOs should align governance responsibilities between CIOs, CISO, and risk management leaders.
Every platform I’ve reviewed offering AI capabilities includes some form of AI governance. But these implementations focus on governing the solution provider’s AI capabilities, whether that means deploying AI agents, authorizing data access, or enabling agent observability.
AI governance capabilities in these platforms are necessary but not sufficient to manage a holistic AI governance strategy, especially for large, regulated enterprises with complex data landscapes.
Collibra, a Leader in the Gartner® Magic Quadrant™ for Data and Analytics Governance Platforms, announced several AI governance solutions at their NYC AI Summit last week. They include several AI governance innovations.
- Collibra Maestro, a studio and assistant for developing AI agents that automate governance tasks.
- Live Map, a real-time context map capturing what entities exist across a company’s documents and how they relate.
- Guardian Agents monitor deployed AI agent behavior and can alert or take corrective action when agents operate outside defined boundaries.
Also notable are capabilities Collibra announced earlier this year, including Collibra’s MCP Server, which brings data definitions and lineage into Claude, ChatGPT, and Databricks. These capabilities matter for organizations looking to grow AI adoption, which is pretty much everyone, and expanding data and AI literacy is a required prerequisite.
CIOs, CDOs, and CISOs looking to implement AI governance, not just policies, should review capabilities that can run ahead of their AI delivery plans.
Leave a Reply