Introduction
The GRC Summit is always a great experience, and the one we held on November 4th and 5th was no exception. It was my second virtual Summit – the first one was in 2021, during Covid – and though we were online, there was no lack of energy. More than 2,000 GRC practitioners, leaders, industry analysts, and risk experts signed up for the experience, and we connected over two mornings to hear how AI is transforming risk, compliance, audit, cyber, and GRC as a whole.
I heard so many terrific quotes: Go further, faster. An agent can't be accountable – a person must be. AI is getting better every day. And one of my favorites – AI is like a toddler. It needs supervision.
Those aren't exactly quotes – they're paraphrased – but they sum up the essence of what we talked about. AI has left pilot and entered production, and the human is still firmly in the loop.
It's hard to summarize everything I heard – I'm still thinking about it more than a week later – but I wanted to share a few key takeaways from the GRC Summit 2026. I hope to see you in person at the London GRC Summit in May!
1. Connect Your Teams and Data So AI Can See the Full Picture
If you were there, you won't forget this image: the flock of birds across the sky (dating myself, like the 80s hair superstars the Flock of Seagulls!)
Our Co-Founder and Vice Chairman Gaurav Kapoor opened with an indelible image of autumnal European skies filled with thousands of birds. They fly in formation for thousands of miles – no committee meeting and no approval chain. When danger appears, the signal moves across the flock, and the flock responds.
Of course, his point was that AI’s biggest value is detecting signals across silos and spotting patterns people can’t see on their own.
An example he gave illustrated this more clearly. A typhoon clogs a port in Taiwan, and an external risk agent flags it. Another agent checks supplier profiles and finds two sole-source suppliers affected. A third finds $25 million of customer demand at risk. Finance and procurement identify contract gaps, and a resilience agent initiates the conversation with the customer. People stay in the loop at each step.
The GRC practitioners speaking at the sessions had their own versions. Saranya Balakrishnan Jothi leads enterprise resilience at First Citizens Bank. She explained how her data became far more useful once risk, resilience, cyber, third-party, and audit teams agreed on what to call a process and a control. Sadhana Deot, Vice President of Global Ethics and Compliance at Trimble, described agents that pull return data from five systems and compile it into a single weekly report. Her team spends less time collecting and more time analyzing.
Both named data quality as the first wall you hit. Saranya had a great quote: with AI, it’s “garbage in, confident garbage out.”
3. Test Any AI Agent on Your Own Hard Cases Before You Trust It
Michael brought the Wizard of Oz to the keynote, and I couldn’t resist this one. When someone says they have agentic AI, ask what information it can access, how it identifies authoritative sources, which actions it can initiate, and where it requires approval. Then test it on your own hard cases: conflicting evidence, an outdated policy, a situation outside its authority. Does it recognize the limit and escalate? In his words: “Just look behind the curtain before you buy the Emerald City.”
He then invited Captain Jack Sparrow to the boardroom. Generative AI is resourceful, persuasive, and remarkably confident. Confidence, he said, is not a dependable measure of the quality of the navigation.
His practical answer is three modes for delegating work to AI. In augment mode, AI assists, and the person decides. In recommend mode, AI compares options, and the person authorizes. In act mode, AI executes within explicit boundaries. These are choices for particular tasks, not a maturity ladder. The rule that ties them together: more consequence and less reversibility call for more human control. And telling an agent in a prompt to behave responsibly is an aspiration, not an access control!
4. Keep Humans in Charge, and Match Oversight to the Stakes
Here's where that toddler quote came in: James Downing, Chief Compliance Officer at JLL Capital Markets, compared AI to a toddler that needs supervision. Solomon Ugah of Rockwell Automation showed us why with the best contrast of the summit. You can watch AI-generated video of Will Smith eating spaghetti and laugh. But when AI gives advice that leads to an explosion in a refinery, “it’s not so funny anymore.”
Hallucination risk depends on the use case, so design the guardrails and the circuit breaker at the start. He also warned that rising trust in AI output can weaken human-in-the-loop checks. His closing instruction: “Watch what AI is doing to you.”
Others on the panels offered practical rules:
Saranya matches explainability to consequence. If a person reviews AI output before anyone acts, she accepts a less transparent model for speed. If AI drives decisions that affect customers, regulatory reporting, or capital, explainability is non-negotiable. Her test is whether she could defend the output to a regulator, and she wants a named owner for each model, not a shared mailbox.
Sadhana ran the same data through AI a week apart and got slightly different results. Her team now checks the calculations with code.
Marc Levine, our CEO, said anything that changes a GRC outcome, such as a risk rating or an audit finding, passes through an approval gate.
Masood Ali of State Street summed up the Day 2 panel: shared intelligence does not mean shared accountability.
5. Boards Want the Forecast and the Proof
Gaurav said boards have stopped asking whether the company is compliant. They now ask whether it is resilient and how quickly it can adapt. Saranya said the question has moved from last quarter to what is expected three, six, and twelve months out. Boards also ask what happens if an AI model fails.
Sadhana’s board no longer wants a static slide. Her team built a distress index that flags upcoming component shortages before production is scheduled, so she arrives with a plan and not just a status.
Then comes the proof. Michael said a count of deployed agents tells you very little. He judges AI on efficiency, effectiveness, resilience, and agility, and he counts review and correction effort alongside time saved. Masood said board conversations are shifting from efficiency to the risk intelligence AI produces.
6. Start With the Problem, and Give Changes 90 Days
I always enjoy the customer stories, because they show what happens after the keynote applause.
Michael Cover’s mantra is “ready, willing, and able.” Does the business area have the foundation, the buy-in, and the time? He starts by defining and quantifying the problem, warns against over-engineering, and waits about 90 days before changing a new process. Feedback at 30 days, he said, often reverses by 60. His outlook on AI: “AI is at its worst today,” which means it’s only getting better.
Jessica Markun, Senior Vice President of Finance and Risk at FCCI Insurance Group, described moving off spreadsheets, documents, and email. Her biggest lesson: “A system requires decisions that a spreadsheet can often just avoid.” She recommends a dedicated business project manager, a decision record, and change messages written by role.
Clifton Roberts, Senior Director of Legal and Business Compliance at Intel, described the direction Intel is working toward in policy governance. If ownership is unclear, he said, a more sophisticated workflow can simply move an unclear decision faster. “A connection nobody can explain is not governance.”
AI in the Platform: A Quick Look at the Product Demos
Two sessions showed the MetricStream platform at work.
Shreyank Kamat’s demo, showed us a compliance manager asking the MetricStream Assistant in plain English whether holding EU prospect data for more than 3 years was within policy. She got an answer with its source, opened a pre-filled issue form, and drafted a control in about five minutes, reviewing each step. In a second demo, AI identified nearly 410 obligations in a good manufacturing practice guide and mapped them to 9 risks, more than 200 controls, and 1 policy.
Chandra Reddy K, our Associate Vice President of Solution Consulting, gave an exclusive first look at our upcoming AI Governance product. One AI use case gets one record that risk, compliance, audit, and cyber teams each see, and a retrained model that failed its fairness checks is automatically blocked from production.
What Comes Next
Michael closed his keynote with Captain Kirk telling his officers that risk is their business. He didn’t want a bridge full of red, amber, and green circles. He wanted to know what changed, what it means, and what needs a decision. That is how GRC gives the business the confidence to boldly go forward.
Our CEO Marc closed the summit by recapping that AI is here to stay, that humans and AI work best together, that outcomes matter more than technology, and that strong foundations come first. He then offered predictions for 2027, including how AI will increasingly show up within the workflows people already use and how AI governance will become core to GRC, as third-party suppliers add AI to their products and agents appear across organizations.
Catch the sessions you missed.
The 2027 edition of the GRC Summit London takes place in person, May 11-12, 2027, at the Royal Garden Hotel in Kensington. I hope to see you there!