Artificial intelligence (AI) can be a powerful workplace tool, but it can also create blind spots. It’s not the technology’s fault, That was the focus of our June 2026 DEI for REAL webinar, where Chris Jones, Rebecca Dessonville, and I explored often-overlooked risks at the intersection of AI, organizational governance, and corporate liability. Chris is a co-founder of Spectra Diversity and its current CEO. Rebecca is founder of Purrly Digital, a human-centered technology consultancy. I, of course, am the founder of Third Ear Conflict Resolution and author of Unsustainable: Why Our Workplaces Aren’t Working and What to Do About It.
Many leaders are being encouraged to view generative AI as a solution for cutting costs and automating knowledge work. But technology cannot replace the human capacity to notice tension, identify inequities, and address conflict before it escalates. When organizations reduce human oversight in favor of algorithmic efficiency, they may also weaken their ability to recognize the workplace friction that costs the U.S. economy an estimated $550 billion annually.
AI can analyze patterns. People experience the impact of those patterns. If leaders want to manage risk effectively, they must continue listening to the humans who see and feel problems first. Inequities rarely begin with a data point. They begin with an experience, and employees will almost always recognize it before the technology does.
Digital Flattery and The Loop
To understand where workplace systems are breaking down, we have to look neutrally at how these predictive engines are built. Rebecca highlighted a stark industry metric from recent market data: the engineering teams within primary Large Language Model (LLM) labs remain significantly homogeneous, with female participation sitting at just 22%—well below the broader tech industry’s 32% baseline.
If we know so little about who is building these systems, what assumptions might be embedded in the systems themselves?
When the workplaces designing our tools lack a wide range of human experiences, those exclusive perspectives are naturally baked into the baseline code. In the tech world, this dynamic is frequently called the Tech Bro Effect. Algorithms struggle to say I don’t know, opting instead to tell you what it seems people generally want to hear (based on the language patterns observed).
At their core, AI platforms are highly advanced text prediction machines built to output the “great middle average” of human data. When leaders rely on generalized AI models to restructure teams, draft policies, or manage delicate human dynamics, they risk creating an echo chamber that rubber-stamps compliance on paper while entirely missing the human realities in the office or on the manufacturing floor.
Compliance is a Weak Defense Against Social Inflation
From a corporate governance perspective, waiting for the law to tell you how to act is a dangerous operational strategy. The legal system is reactive. Its primary function is to correct past harms after an individual has already suffered personal or property injury.
Some states are beginning to mandate disclosures if AI is utilized in resume screening or virtual hiring interviews, but the traditional plaintiffs’ bar is moving much faster than official statutory updates. They are actively leveraging traditional anti-discrimination frameworks, like disparate impact metrics, to challenge automated personnel decisions. The Equal Employment Opportunity Commission (EEOC) might have deprioritized disparate impact claims, but employers are still accountable to the states in which they have employees (or workers who could be deemed their employees).
As I mention in my upcoming book, Unsustainable, a less regulated environment does not necessarily equal less liability. When formal regulatory bars are lowered, public expectations invariably rise to fill the vacuum. This structural disconnect is fueling massive “social inflation verdicts” delivered by juries composed of anxious, anti-corporate citizens. True risk management requires looking beyond basic checkboxes. We must cultivate an internal operating standard built on mutual respect and functional equity.
Activating the HAQ Framework
Resolving workplace friction requires establishing systems that ensure every decision is Humane, Affordable, and Quick (HAQ). If you choose to adopt AI as an operational partner, you must actively engineer it to meet those standards. Here are three practical ways to protect your organization’s legacy and bottom line, or HAQ them:
-
Re-Engineer Your Operational Prompts. Move away from generic, generalized commands. Explicitly program human-centric and accessible parameters into your instructions. For example: “Draft a team calibration message utilizing gender-neutral and accessible language. Avoid default structural assumptions regarding household dynamics and utilize singular ‘they’ pronouns.”
-
Audit Tech Outputs for Hidden Variances. Never treat AI drafts as finished products. Run generative text back through alternative filters to scan for structural blind spots. Command the system: “Review this policy draft for exclusionary corporate jargon or implicit ageist terminology, highlight variances, and suggest alternative language built on clarity.”
-
Enforce an Uncompromising “Human in the Loop” Rule. Technology excels at processing flight data, but humans must retain the ultimate decision-making power because AI lacks emotional intelligence, passion, and context. Don’t allow an unmoderated algorithm to dictate client communication or final personnel evaluations without deep, empathetic human oversight.
If you understand AI as a series of ones and zeros rather than an independent assistant you don’t need to supervise, you are less likely to make a major legal misstep. Remember that AI output is purely a reflection of its training data. This means it requires a HITL to keep it fair, accurate, and effective. That comes with its own risks, but they are more manageable than a “black box.”