Implementation

GRC orchestration with AI: workflows, SLAs and approvals across teams

· 12 min read ·

Orchestrating GRC means getting a finding to the end of its life cycle without it getting lost between teams. A vulnerability found by security passes through the risk function, the business process owner, the IT team that applies the fix, compliance and, later, internal audit. Every hand-off is a point where the item can end up without an owner, a deadline or proof. Orchestration defines those hand-offs; AI helps keep them moving; the decision stays with whoever has the authority to make it.

This article is not about which tasks to automate first or how to turn controls into recurring tests. Those topics are covered in AI compliance automation: where to start and Continuous internal audit: how to automate it. The focus here is what happens between teams: roles, hand-offs, deadlines, escalation, integration with the ticketing tool and the approval trail.

Where GRC workflows break: at the hand-offs

Within each team, work usually moves. Security knows how to run a scanner, IT knows how to apply a fix, audit knows how to test a control. What fails is the hand-off from one team to the next. The symptoms repeat themselves:

  • A finding with no owner. The penetration test report arrives by email, becomes a spreadsheet and nobody knows which team is accountable for each line.
  • Treatment with no decision. The fix is scheduled, but nobody in the business decided whether the risk would be mitigated, accepted or transferred.
  • An exception with no expiry. A deadline is extended "until next quarter" in a chat thread and never reviewed again.
  • Closure with no verification. The ticket was closed, but nobody confirmed that the weakness is gone.
  • An audit with no trail. When the auditor asks who accepted the risk and on what basis, the answer is scattered across email, chat and meeting minutes.

None of these problems is solved by making a task faster. They are solved when the workflow is explicit: every step with an input, an output, an owner, a deadline and a rule for passing to the next one.

Who takes part and which decision belongs to whom

Before drawing steps, define roles. The NIST Cybersecurity Framework 2.0 asks for roles, responsibilities and authorities related to cybersecurity risk management to be established, communicated, understood and enforced (GV.RR-02). The IIA's Three Lines Model, updated in July 2026, helps separate who operates, who oversees and who provides independent assurance. A common split, which you should adapt to your structure:

TeamRole in the workflowDecision that belongs to it
SecurityLogs the finding, assesses technical severity and verifies the fix.Technical classification and re-test result.
Risk owner (business)Accountable for the affected process or asset.Mitigate, accept, transfer or avoid, within their authority.
IT / system ownerCarries out the treatment plan.How and when to implement, within the approved deadline.
Risk (second line)Links the finding to the risk register, challenges the assessment and tracks deadlines.Whether the decision is within appetite and needs to go up a level.
ComplianceLinks the item to the obligations and frameworks affected.Whether there is a regulatory or contractual obligation to report or meet.
Internal audit (third line)Independently tests whether the workflow worked.Assurance opinion; it does not own the plan.

The most sensitive point is the third column. Security and risk recommend and challenge, but the one who accepts a risk is the part of the business that lives with it. And audit must stay out of execution to preserve its independence.

An end-to-end workflow example: from finding to verification

Hypothetical example. A penetration test finds an authentication flaw in the customer portal, an internet-facing system that processes personal data. The deadlines below are illustrative; yours should come from your vulnerability management policy and your risk appetite.

StepWho is accountableWhat the step producesDeadline (illustrative)
1. Log the findingSecurityA single finding (no duplicates), with asset, technical evidence and severity.1 working day
2. Link to riskRisk, with security and complianceFinding linked to a risk scenario, to the process owner and to the obligations affected (here, the security measures required by Article 46 of the LGPD, Brazil's General Data Protection Law).2 working days
3. Treatment decisionRisk ownerRecorded decision: mitigate, accept, transfer or avoid, with a rationale.3 working days
4. Plan and ticketIT / system ownerPlan with milestones and a ticket opened in the ITSM tool, linked to the risk.Same day as the decision
5. Exception, if the deadline cannot be metRisk owner, within the authority set by the appetiteException with a compensating control, an expiry date and a named approver.Before the fix deadline expires
6. Execution and evidenceITFix applied and evidence attached to the ticket.15 calendar days for a critical flaw in an exposed system
7. VerificationSecurity (re-test) and risk (review)Re-test result; residual risk reassessed.5 working days after the ticket is closed
8. Closure and reportingRiskItem closed with a complete trail; indicators updated. Audit may sample it later.Next reporting cycle

Three rules make this workflow work. First, no step starts without the output of the previous one: there is no ticket without a treatment decision. Second, closing the ticket at step 6 closes nothing; it triggers step 7. Third, if step 3 results in acceptance, the workflow moves to a formal acceptance, with an expiry and a review, not simply to the archive.

Servers in a modern data centre representing an organisation's risk and compliance infrastructure

SLAs and escalation: deadlines that depend on risk

A single SLA for everything ("30 days to fix") fails both ways: it is too slow for what is urgent and it generates exceptions in bulk for what is not. A useful reference is CISA's Binding Operational Directive 26-04, of 10 June 2026, which applies to US federal civilian agencies. It supersedes BOD 22-01 and makes remediation urgency depend on four variables: whether the asset is publicly exposed, whether the vulnerability is in the KEV catalogue, whether exploitation can be automated and what the technical impact is. The directive also requires policies to assign roles and responsibilities and to establish internal validation, tracking and reporting. It does not apply to Brazilian companies, but the design is a good model. The logic of prioritising by exploitation and exposure is covered in 48,000 CVEs a year: where to start.

In practice, three design decisions prevent most delays:

  1. A clock per step, not just overall. Measuring only the final deadline hides where the item stalled. If the treatment decision takes two weeks, the problem is not IT's.
  2. An escalation ladder agreed before anything is late. For example: a reminder to the owner when part of the deadline remains; a notice to their manager on the due date; a notification to the risk owner after a fixed period; and an agenda item at the risk committee for overdue critical items. Each rung needs a name, not a mailing list.
  3. The clock only stops with a decision. Pausing an SLA requires an approved exception with an expiry. "Waiting for the supplier" with no record pauses nothing.

Exceptions and risk acceptance: where the trail matters most

Exceptions are where the audit trail usually fails, because they start as conversations. NIST CSF 2.0 addresses this explicitly: risk responses are chosen, prioritised, planned, tracked and communicated (ID.RA-06), and changes and exceptions are managed, assessed for risk impact, recorded and tracked (ID.RA-07). ISO/IEC 27001 requires risk owners to approve the treatment plan and accept the residual risk (clause 6.1.3), as detailed in ISO 27001: a step-by-step guide to certification.

An exception that stands up to audit has, at a minimum:

  • A rationale for why the deadline or control cannot be met.
  • A compensating control and who checks that it is in place.
  • An expiry date, with a mandatory review before it lapses, never automatic renewal.
  • An approver with the right authority, set by the level of risk. The higher the residual risk, the more senior the approver; the limits come from risk appetite translated into numbers.
  • Reopening triggers, such as the vulnerability being added to the KEV catalogue or a change to the asset.

For the treatment plan, the plan of action and milestones (POA&M) format in NIST SP 800-53 is a good reference: control CA-5 calls for a plan that documents planned remediation actions and is updated based on control assessments, independent audits and continuous monitoring.

Integration with ticketing and ITSM: which system records what

Most companies already have a ticketing tool where IT works, and the GRC workflow should not compete with it. The split that usually works is by system of record, whether that is a GRC platform, a set of specialist tools or the ITSM tool itself with a risk module. What matters is that each type of record has a single source of truth:

  • GRC record: risk, treatment decision, exception, approval, evidence and verification.
  • ITSM: execution of work, the technical team's queue, change and maintenance window.

The integration between the two needs to sync, in both directions, at least the risk identifier, the approved deadline, the status and the attached evidence. Two precautions avoid most problems. The deadline should not be editable only on the ITSM side, because changing the deadline is an exception and requires approval. And the "resolved" status in the ITSM tool should open the verification step in the GRC record, instead of closing the risk. Approvals by email or chat sit outside the trail; if they happen, they need to be recorded in the workflow.

At the hand-offs: where AI helps and where a person decides

In a cross-team workflow, AI adds more value in getting the item to the right person with the right context than in the decision itself. Four hand-offs from the example illustrate the split; autonomy levels by type of action are covered in AI compliance automation.

AI canA person decides
Suggest the owner and route the item.Who owns the risk when accountability is disputed.
Summarise the history and treatment options for the approver.Mitigate, accept, transfer or avoid.
Warn when a deadline is likely to be missed and when an exception is about to expire.Escalate, renew or refuse the exception.
Assemble the evidence pack for verification and audit.Declare the item verified and closed.

Every AI suggestion that is accepted or rejected should stay in the item's history, with who decided and when. Without that, the trail shows the outcome but not who made the decision. Governance of the models themselves is the subject of AI Governance Software: What It Is and How to Choose It.

How to tell whether orchestration is working

A few indicators, measured per step, show where the workflow stalls:

  • Average time in each step, especially between logging and the treatment decision.
  • Percentage of items within deadline, by risk level.
  • Exceptions that expired without review.
  • Items reopened after verification, a sign of premature closure.
  • Findings with no owner for more than a few days.

The ones that matter to leadership can become key risk indicators, as described in KRIs: key risk indicators every CISO should monitor.

Where Eagle fits

Eagle, Futture's Enterprise Risk Intelligence platform, was designed to connect this chain in a common model, from asset and exposure to the action plan that treats risk traceably, and on to the executive decision. The site describes an audit trail with a history of changes and decisions and role-based permissions, which support approval authorities. The list of integrations includes ticketing tools, with the caveat that the availability of each connector depends on the environment and the contracted scope. Eagle's AI supports classification, prioritisation and recommendations in a human-in-the-loop model: the user reviews, approves, rejects or adjusts every relevant result.

Conclusion

Pick a workflow that gets lost between teams today, such as handling penetration test findings, and draw it on a single page with the eight steps of the example. For each hand-off, write down who delivers, who receives, the deadline and the rule for passing it on. Then pull the last 20 items in that workflow and measure how long each one spent in each step. The bottleneck usually shows up in the treatment decision or in verification, and that is where it pays to start orchestrating, before automating anything else.

Further reading: AI compliance automation: where to start, Continuous internal audit: how to automate it and Risk appetite: from sentence to number.

Frequently asked questions

What is the difference between GRC orchestration and GRC automation?

Automation speeds up a task, such as collecting a piece of evidence or testing a control. Orchestration coordinates the whole workflow across teams: who receives each item, in what order, by when, who approves it and how closure is verified. You can automate tasks without orchestrating, but then items still get lost between one team and the next.

Who should approve an exception or a risk acceptance?

The risk owner, meaning whoever is accountable for the affected process or asset, within the approval authority set by the risk appetite policy. ISO/IEC 27001 requires risk owners to approve the treatment plan and accept the residual risk. Security and the risk function recommend and challenge, but they do not accept risk on the business's behalf.

Does closing the ticket in the ITSM tool close the risk?

No. The ticket records that the work was carried out. The risk should only be closed after an independent check of that work, such as a re-test or a second-line review of the evidence. That is why closing the ticket should trigger verification, not closure.

At which points in the workflow should AI not decide on its own?

Accepting risk, approving exceptions, setting priorities when resources do not cover everything, and closing an item as verified. At those points AI can prepare the context, suggest options and flag deadlines, but the decision belongs to a person with the authority to make it, and it is recorded.

Official references

About the author

Founder, Futture.ai | CISSP

Cybersecurity, governance and risk management specialist with over 25 years of industry experience. CISSP-certified, founder of Network Secure and Futture.ai.

Read next

← Back to the blog