E-Paper
AI in the boardroom: Leadership accountability challenge
This article is authored by Agamjeet Dang, CEO, Executive Access India.
Every cricket fan knows the feeling. The ball-tracking graphic comes up on the big screen, and for half a second, everyone in the stadium treats it like gospel. It isn't. It's a projection. It’s an educated guess rendered beautifully enough to look like a fact. Cricket knew this before anyone else did, which is why it built a rule for the doubt. The umpire's call. When the ball only grazes the stumps, the technology steps back, and the human decision on the field stands. The umpire's name goes into the scorecard. Not the machine's.
I think about this rule a lot in boardrooms, because we've enthusiastically adopted the projection and quietly dropped the doubt.
AI didn't enter our companies the way every other technology has--through a budget line, a pilot, or a rollout plan we could track. It directly got a seat at the table. It recommends, filters, approves, and increasingly acts on its own. It decides who gets shortlisted, which transaction gets flagged, what a customer sees the second an app opens. And most Indian boards still file it under technology, wedged somewhere between cybersecurity and vendor contracts.
That's a mistake. AI isn't a tool the company uses anymore. It's becoming the system through which the company decides. Traditional governance assumes a person made the call, inside a structure you can point to. AI quietly breaks that assumption. The board isn't just overseeing people and processes now. It's overseeing a system it often can't see into, and if we're being honest, can't always explain either.
Here's the trap, and it's a comfortable one. Boards don't feel like they've lost control. AI output arrives clean, confident, and quantified. It looks like logic, not instinct. That's exactly why nobody questions it.
I've watched this play out. A lender rolled out an AI credit model to approve loans faster. Approvals climbed, portfolio quality held, and nobody had a reason to dig further. Quietly, the model started excluding entire customer segments that just didn't resemble the patterns it trusted. Every number the board saw was defensible. The outcome wasn't. And nobody had done anything wrong. The board had approved the rollout, reviewed the dashboards, every quarter, on time. What it never asked was how the model was deciding. That's not a performance failure. That's a judgement failure, and it slipped through because it looked like success.
Here's the uncomfortable truth most boards sense but rarely articulate. Management understands these systems far better than the board does. Directors are fluent in capital and risk. Model behaviour is a foreign language most of us never enrolled to learn. So, authority drifts, quietly, without anyone deciding to hand it over. We accept explanations instead of interrogating them, because we don't yet know what to ask. The board is still in the room. It's just not always steering.
Oversight assumes the machine is sound and someone's watching. Accountability asks the harder question--who owns this decision, and who answers when it's wrong? This isn't about turning directors into engineers. It's about knowing enough to ask the right question before the post-mortem, not during it.
Come to think of it, what are we optimising for? This is not a technical question. It's a governance one, and it belongs at the main table, next to succession and capital allocation, not as a footnote beneath them. Ethics can't be the last slide before lunch either. A system reflects exactly what it's trained on and told to chase. Build ethics into the design and the measurement, or it isn't really there. It's just a decoration.
The cost I find hardest to say out loud: I've watched experienced bankers stop questioning an automated risk engine simply because it had been right, at scale, for years. Behaviour they'd once flagged in a heartbeat stopped looking unusual, because the system had stopped flagging it. The danger isn't that technology fails. It's that we quietly stop thinking past it.
That makes the board custodian more than the system. Perhaps it's a custodian of thinking itself. An institution that transforms without protecting its own judgement ends up performing brilliantly while reasoning poorly. That's a dangerous kind of success.
Regulators will show up eventually, late and unforgiving, and they won't ask whether we adopted AI. They'll ask how responsibly we governed it.
(The views expressed are personal)
This article is authored by Agamjeet Dang, CEO, Executive Access India.
