How are you governing your AI risks today?
5-minute read
Three numbers show where AI stands in risk and compliance today:
More use - 53% of risk and compliance professionals now use AI or are testing it. This is up from 30% in 2023 (Moody’s, survey of 600 professionals, January 2026).
More independence - Almost one in three new AI projects at the 50 large banks tracked by Evident in the first quarter of 2026 used agentic AI. This means the system takes action instead of only giving advice.
Little proof - Only 30% of risk and compliance professionals say AI has delivered the benefits they expected. Also, only 27% of new bank AI projects announced in the second quarter of 2026 had any reported results (Evident).
In short, AI use is growing quickly, and AI systems are starting to act on their own. Yet fewer than one in three users can show that AI has delivered the benefits they expected. This creates a challenge for chief risk officers, compliance officers and operations leaders. Their executive committees, auditors and regulators will ask three questions: Where is AI being used? How is it being controlled? And what results has it delivered? So far, few companies can answer all three questions with evidence.
Risk management must stay ahead of regulation
Two lessons have stayed with me throughout my career: expect the unexpected, and never assume you have prepared for everything. AI tests both lessons. It is already being used in approved tools and in tools that employees have chosen themselves. Few chief risk officers can see all the places where it is being used. You must be able to trust the information going in, the way it is processed and the results it produces, even when you cannot always see how the system works inside. Regulators will eventually move towards common rules, but risk management must stay ahead of regulation, not follow it. I have never managed risk just to meet regulatory requirements. You must follow the rules, but if something goes badly wrong, you cannot blame the regulator. You are responsible. Start now, because serious mistakes happen when you are trying to catch up.
Grigorios (Greg) Markouizos is a former Citi executive with more than 30 years of banking experience. He has chaired Euroclear’s Collateral User Group and served on industry committees covering regulation, market infrastructure and risk.
Why time is running out
Regulators are not waiting for one set of rules that applies everywhere. Reuters reported in June that the OCC and the Federal Reserve now ask about AI during regular examinations. They ask about human checks, emergency shutdown switches and plans for leaving a technology supplier. This is happening even though SR 26-2 does not include generative AI or agentic AI in its formal guidance on model risk. On 7 October 2026, the Monetary Authority of Singapore (MAS) issued its final Guidelines on AI Risk Management. Requirements for oversight, AI inventories and assessments of how important each risk is will apply from October 2027. Controls covering the full life of AI systems will apply from October 2028. In the EU, rules for high-risk AI, including credit scoring, will apply from 2 December 2027. Companies that start now will have a year of evidence ready when regulators ask questions.
Figure 1: Important dates for regulators and the actions companies need to take
What GALE is
GALE stands for ICAG Partners’ GRC AI Leverage Enhancement framework. It helps companies use AI in governance, risk and compliance in ways they can measure and support with evidence.
Figure 2: A quick look at GALE
GALE is based on the ICAG GRC Excellence Model (IGEM©). This model divides governance, risk and compliance (GRC) into 32 areas across three levels. The first is the organisation's foundation, which includes strategy, people, data, AI, security and governance. The second covers the work that delivers value, from checking compliance to testing systems by acting like an attacker. The third is a risk and compliance cycle that covers everything from regulations to reports sent to regulators.
Figure 3: The ICAG GRC Excellence Model (IGEM©)
Each area is given a score using a five-level maturity model called IGMM©. The levels range from Initial to Optimised. For example, in compliance assurance, Level 1 means people carry out manual checks from time to time and collect evidence afterwards. At Level 3, agentic AI checks controls all the time, records every action taken by each AI agent, and has a person approve the results. At Level 5, groups of AI agents fix low-risk problems themselves, while compliance staff focus on decisions that need human judgement. Executive committees receive one score for each area. They can track these scores and compare them with others.
Figure 4: The IGMM© model for compliance assurance, from Initial to Optimised
Six things that make it work
Set a starting point for your AI governance and AI use in three weeks. Begin with our free online self-assessment (see here), then take part in working sessions with our experts.
Move from checks every three months to continuous assurance with Perpetual GRC. AI agents carry out compliance, risk and control checks around the clock. Your teams review the results and approve them.
Use automation to design and test controls. This includes using AI to draft control descriptions and carrying out ongoing tests through APIs.
Create a list of all your AI systems, group them by risk level and set up controls that follow the NIST AI RMF, ISO/IEC 42001 and the new MAS guidelines. This helps examiners see a properly managed set of AI systems instead of just a list of test projects.
Bring together rules that apply in the US, UK, EU, Middle East and Asia-Pacific in one register. Link these rules to your policies and controls.
Keep improving through quarterly maturity checks, AI red-team tests, planning for possible situations and reports that help your executive committee take action.
The work follows seven steps, starting with an assessment and ending with ongoing improvement. Quick improvements happen alongside the assessment, so companies can see results within weeks. We do not have a financial interest in any GRC or AI governance platform. This means we recommend tools based on how well they fit your existing systems.
Figure 5: The seven steps in the GALE approach
Where do you stand?
Our free AI GRC Maturity Check takes about five minutes. It shows how mature your current approach is, which gaps matter most and what you can do most quickly to reach the next level. Our team can also provide a detailed report covering each area and a comparison with other companies in your field.
Pilot projects do not prove that a company will pass an audit. Evidence does. Your score is the first piece of that evidence.
About ICAG Partners
ICAG Partners is an independent advisory firm. Its experts have built their careers in risk, compliance and audit. The firm helps banks, capital markets companies, asset and investment managers, and broker-dealers assess, design and deliver AI-powered governance, risk and compliance services. This work includes maturity checks, AI governance, ongoing compliance and automated control checks. Full details of its solutions are available at https://icagpartners.com/
About the Authors
CEO, ICAG Partners
Subas Roy is the Founder and CEO of ICAG. He has held senior executive positions at other major companies for more than 30 years. He advises and works closely with large and mid-sized private equity firms and their operating partners. His work helps them create more value, increase company market values, prepare for sales and find answers to difficult questions about technology and AI, data, risk, people and processes.
He has extensive experience in the UK and global financial services industry, with a special focus on banking and insurance.
Partner, ICAG Partners
Gopal Kuchibhotla is a Partner and Head of Risk Advisory at ICAG. He has more than 35 years of experience advising organisations around the world and across different industries. His work covers risk, finance and technology. He advises private equity firms and their operating partners on how to create, protect and realise value across the companies they own.
His areas of expertise include risk and compliance, technology, AI and data, and mergers and acquisitions. This includes due diligence, combining companies after a merger and designing the way an organisation operates.
