As companies incorporate cybersecurity into governance, threat and compliance (GRC), it is very important revisit current GRC packages to make sure that the rising use and dangers of generative and agentic AI are addressed so companies proceed to satisfy regulatory necessities.
“[AI] It’s a massively disruptive know-how in that it’s not one thing you may put right into a field and say ‘properly that’s AI’,” says Jamie Norton, member of the ISACA board of administrators and CISO with the Australian Securities and Funding Fee (ASIC).
It’s arduous to quantify AI threat, however knowledge as to how the adoption of AI expands and transforms a company’s threat floor gives a clue. In keeping with Verify Level’s 2025 AI security report, 1 in each 80 prompts (1.25%) despatched to generative AI providers from enterprise units had a excessive threat of delicate knowledge leakage.
CISOs have the problem to maintain tempo with enterprise calls for for innovation whereas securing AI deployments with these dangers in view. “With their pure security hat on, they’re making an attempt to cease shadow AI from turning into a cultural factor the place we are able to simply undertake and use it [without guardrails],” Norton tells CSO.
AI is just not a typical threat, so how do GRC frameworks assist?
Governance, threat and compliance is an idea that originated with the Open Compliance and Ethics Group (OCEG) within the early 2000s as a approach to outline a set of essential capabilities to deal with uncertainty, act with integrity, and guarantee compliance to help organizational targets. Since then, GRC has developed from guidelines and checklists targeted on compliance to a broader method of managing threat. Data safety necessities, the rising regulatory panorama, digital transformation efforts, and board-level focus have pushed this shift in GRC.
On the identical time, cybersecurity has develop into a core enterprise threat and CISOs have helped guarantee compliance with regulatory necessities and set up efficient governance frameworks. Now as AI expands, there’s a necessity to include this new class of threat into GRC frameworks.
Nevertheless, trade surveys recommend there’s nonetheless an extended approach to go for the guardrails to meet up with AI. Solely 24% of organizations have totally enforced enterprise AI GRC insurance policies, in accordance with the 2025 Lenovo CIO playbook. On the identical time, AI governance and compliance is the primary precedence, the report discovered.
The trade analysis means that CISOs might want to assist strengthen AI threat administration as a matter of urgency, pushed by management’s starvation to understand some pay-off with out transferring the danger dial.
CISOs are in a troublesome spot as a result of they’ve a twin mandate to extend productiveness and leverage this highly effective rising know-how, whereas nonetheless sustaining governance, threat and compliance obligations, in accordance with Wealthy Marcus, CISO at AuditBoard. “They’re being requested to leverage AI or assist speed up the adoption of AI in organizations to realize productiveness positive factors. However don’t let or not it’s one thing that kills the enterprise if we do it incorrect,” says Marcus.
To help risk-aware adoption of AI, Marcus’ recommendation is for CISOs to keep away from going alone and foster broad belief and buy-in to threat administration throughout the group. “The actually vital factor to achieve success with managing AI threat is to method the state of affairs with a collaborative mindset and broadcast the message to of us that we’re all in it collectively and also you’re not right here to gradual them down.”
This method ought to assist encourage transparency about how and the place AI is getting used throughout the group. Cybersecurity leaders should attempt to get visibility by establishing a security course of operationally that can seize the place AI’s getting used at present or the place there’s an rising request for brand new AI, says Norton.
“Each single product you’ve bought as of late has some AI and there’s not one governance discussion board that’s selecting all of it up throughout the spectrum of various kinds [of AI],” he says.
Norton suggests CISOs develop strategic and tactical approaches to outline the several types of AI instruments, seize the relative dangers, and steadiness potential pay-off in productiveness and innovation. Tactical measures similar to safe by design processes, IT change processes, shadow AI discovery packages or risk-based AI stock and classification are sensible methods to cope with the smaller AI instruments. “The place you will have extra day-to-day AI — that little bit of AI sitting in some product or some SaaS platform, which is rising in every single place — this is likely to be managed by way of a tactical method that identifies what [elements] want oversight,” Norton says.
The strategic method applies to the massive AI adjustments which might be coming with main instruments similar to Microsoft Copilot and ChatGPT. Securing these ‘huge ticket’ AI instruments utilizing inner AI oversight boards is considerably simpler than securing the plethora of different instruments which might be including AI.
CISOs can then focus their sources on the highest-impact dangers in a method that doesn’t create processes which might be unwieldy or unworkable. “The concept is to not bathroom this down in order that it’s nearly inconceivable to get something, as a result of organizations sometimes wish to transfer rapidly. So, it’s extra of a comparatively light-weight course of that applies this consideration [of risk] to both permit AI or be used to stop it if it’s dangerous,” Norton says.
Finally, the duty is for security leaders to use a security lens to AI utilizing governance and threat as a part of the broader GRC framework within the group. “A whole lot of organizations can have a chief threat officer or somebody of that nature who owns the broader threat throughout the surroundings, however security ought to have a seat on the desk,” Norton says. “Nowadays, it’s not about CISOs saying ‘sure’ or ‘no’. It’s extra about us offering visibility of the dangers concerned in doing sure issues after which permitting the group and the senior executives to make choices round these dangers.”
Adapting current frameworks with AI threat controls
AI dangers embody knowledge security, misuse of AI instruments, privateness issues, shadow AI, bias and moral issues, hallucinations and validating outcomes, authorized and reputational points, and mannequin governance to call a number of.
AI-related dangers needs to be established as a definite class throughout the group’s threat portfolio by integrating into GRC pillars, says Dan Karpati, VP of AI applied sciences at Verify Level. Karpati suggests 4 pillars:
- Enterprise threat administration defines AI threat urge for food and establishes an AI governance committee.
- Mannequin threat administration screens mannequin drift, bias and adversarial testing.
- Operational threat administration contains contingency plans for AI failures and human oversight coaching.
- IT threat administration contains common audits, compliance checks for AI programs, governance frameworks and aligning with enterprise targets.
To assist map these dangers, CISOs can take a look at the NIST AI Danger Administration Framework and different frameworks, similar to COSO and COBIT, and apply their core rules — governance, management, and threat alignment — to cowl AI traits similar to probabilistic output, knowledge dependency, opacity in resolution making, autonomy, and fast evolution. An rising benchmark, ISO/IEC 42001 gives a structured framework for AI for oversight and assurance that’s supposed to embed governance and threat practices throughout the AI lifecycle.
Adapting these frameworks affords a approach to elevate AI threat dialogue, align AI threat urge for food with the group’s overarching threat tolerance, and embed strong AI governance throughout all enterprise items. “As an alternative of reinventing the wheel, security leaders can map AI dangers to tangible enterprise impacts,” says Karpati.
AI dangers will also be mapped to the potential for monetary losses from fraud or flawed decision-making, reputational harm from data breaches, biased outcomes or buyer dissatisfaction, operational disruption from poor integration with legacy programs and system failures, and authorized and regulatory penalties. CISOs can make the most of frameworks like FAIR (issue evaluation of knowledge threat) to evaluate the probability of an AI-related occasion, estimate loss in financial phrases, and entry threat publicity metric. “By analyzing dangers from each qualitative and quantitative views, enterprise leaders can higher perceive and weigh security dangers in opposition to monetary benchmarks,” says Karpati.
As well as, with rising regulatory necessities, CISOs might want to monitor draft laws, observe requests for remark durations, have early warnings about new requirements, after which put together for implementation earlier than ratification, says Marcus.
Tapping into trade networks and friends can assist CISOs keep throughout threats and dangers as they occur, whereas reporting capabilities in GRC platforms monitor any regulatory adjustments. “It’s useful to know what dangers are manifesting within the discipline, what would have protected different organizations, and collectively constructing key controls and procedures that can make us as an trade extra resilient to all these threats over time,” Marcus says.
Governance is a essential a part of the broader GRC framework and CISOs have an vital position in setting the organisational guidelines and rules for a way AI is used responsibly.
Creating governance insurance policies
Along with defining dangers and managing compliance, CISOs are having to develop new governance insurance policies. “Efficient governance wants to incorporate acceptable use insurance policies for AI,” says Marcus. “One of many early outputs of an evaluation course of ought to outline the principles of the street to your group.”
Marcus suggests a stoplight system — purple, yellow, inexperienced — that classifies AI instruments to be used, or not, throughout the enterprise. It gives clear steerage to staff, permits technically curious staff a protected area to discover whereas enabling security groups to construct detection and enforcement packages. Importantly, it additionally let security groups provide a collaborative method to innovation.
‘Inexperienced’ instruments have been reviewed and accredited, ‘yellow’ require extra evaluation and particular use instances, and people labelled ‘purple’ lack the required protections and are prohibited from worker use.
At AuditBoard, Marcus and the crew have developed a regular for AI device choice that features defending proprietary knowledge and retaining possession of all inputs and outputs amongst different issues. “As a enterprise, you can begin to develop the requirements you care about and use these as a yardstick to measure any new instruments or use instances that get offered to you.”
He recommends CISOs and their groups outline the guiding rules up entrance, educate the corporate about what’s vital and assist groups self-enforce by filtering out issues that don’t meet that normal. “Then by the point [an AI tool] will get to the CISO, folks have an understanding of what the expectations are,” Marcus says.
In relation to particular AI instruments and use instances, Marcus and the crew have developed ‘mannequin playing cards’, one-page paperwork that define the AI system structure together with inputs, outputs, knowledge flows, supposed use case, third events, and the way the info for the system is educated. “It permits our threat analysts to judge whether or not that use case violates any privateness legal guidelines or necessities, any security finest practices and any of the rising regulatory frameworks which may apply to the enterprise,” he tells CSO.
The method is meant to establish potential dangers and be capable of talk these to stakeholders throughout the group, together with the board. “If you happen to’ve evaluated dozens of those use instances, you may pick the frequent dangers and customary themes, combination these after which give you methods to mitigate a few of these dangers,” he says.
The crew can then take a look at what compensating controls may be utilized, how far they are often utilized throughout totally different AI instruments and supply this steerage to the manager. “It shifts the dialog from a extra tactical dialog about this one use case or this one threat to extra of a strategic plan for coping with the ‘AI dangers’ in your group,” Marcus says.
Jamie Norton warns that now the shiny interface on AI is quickly accessible to everybody, security groups want to coach their concentrate on what’s taking place underneath the floor of those instruments. Making use of strategic threat evaluation, using threat administration frameworks, monitoring compliance, and growing governance insurance policies can assist CISOs information the group in its AI journey.
“As CISOs, we don’t wish to get in the way in which of innovation, however we’ve got to place guardrails round it in order that we’re not charging off into the wilderness and our knowledge is leaking out,” says Norton.
