Risk treatment, monitoring and review, and incident response

SDAIA's National AI Risk Management Framework (SDAIA-P145) is an advisory framework with four reference pillars and five steps. The last two steps are risk treatment «معالجة المخاطر» (p. 26–27) and monitoring and review «المتابعة والمراجعة» (p. 28–29). The example section also gives risk treatment examples (p. 42–44). For the framework and its structure, see National AI Risk Management Framework and Framework structure.

Treatment strategy selection

The risk treatment step of SDAIA-P145 selects and applies the applicable risk treatment for each risk of an AI system (p. 26). The selection uses the risk level from the risk assessment and the context of use of the system. In SDAIA-P145, the step makes sure that the residual risk decreases to an acceptable risk level before release or during operation.

In SDAIA-P145, treatment options are in proportion to the severity, type, and scope of impact of the risk (p. 26). The first part of the risk treatment section gives these examples of treatment options (p. 26):

• Mitigation with technical controls, organizational controls, and operational controls
• Acceptance of a risk of small impact that agrees with the approved risk tolerance
• Escalation of the risk to the competent parties for the applicable decision
• Avoidance when the risk is high: a change to the design, a limit on use, or a stop of all or part of the system

The acceptance treatment option below gives its conditions.

In SDAIA-P145, it is necessary in this step to record the risk treatment decisions, the acceptance decisions, and the escalation decisions with a method (p. 26). The records identify the responsibilities of each person or party, the approval mechanisms, and the follow-up indicators. They also connect the decisions to the governance of the system lifecycle. The step has four tasks: the treatment strategy task, control design and implementation, residual-risk reassessment, and the acceptance or escalation decision.

The treatment strategy task «تحديد استراتيجية المعالجة لكل خطر» selects the most applicable risk treatment for each risk, from the risk level and the context of use (p. 26). The risk treatment uses controls and measures in proportion to the severity of the effects and to their likelihood. SDAIA-P145 gives this sequence for the task:

1. Examine the risk characterization and the results of the risk assessment (the risk level).
2. Identify the causes of the risk, when it occurs, how it occurs, and the affected parties.
3. Set the treatment objective clearly. The treatment objective is one of these:
   • Prevent the risk.
   • Decrease its likelihood or impact.
   • Transfer part of its effects.
   • Accept the risk in the approved risk tolerance limits.
4. Select one of the four treatment options. Each selection has a decision justification, and an audit can examine it.

SDAIA-P145 gives these four treatment options (p. 26):

Treatment option	Treatment option in SDAIA-P145 (p. 26)
Avoidance	A change to the system design, a change to the use cases, a limit on functions, or a stop of functions. Avoidance applies when the risk is high and it is not possible to decrease the risk to an acceptable risk level.
Mitigation	Technical controls, organizational controls, and operational controls that decrease likelihood and impact. Examples are an increase in data quality, a decrease in bias, and an increase in security and robustness. Other examples are control of permissions, an increase in transparency, added human intervention, and controls that monitor deviations.
Risk transfer	Part of the risks or of their effects goes to a different party through contractual arrangements, guarantees, insurance, or obligations on suppliers. SDAIA-P145 tells entities that risk transfer does not cancel the entity's governance and compliance responsibility.
Acceptance	Acceptance applies when the residual risk is shown to be in the approved tolerance level, after a check shows that the minimum controls are available. The acceptance record gives the decision justification, the limits of use, the acceptance conditions, and the review plan.

In SDAIA-P145, it is necessary in this task to record the decision at the level of the risk register (p. 26). The record identifies these items:

• The decision justification.
• The responsibilities of each person or party.
• The approval and escalation requirements for high risks.
• The connection between the treatment strategy and measurement and follow-up indicators. They check that risk treatment works and continues to be applicable during operation.

In SDAIA-P145, this record makes it possible to apply and to examine the risk treatment, with accountability. For the approved risk tolerance and the decision rights of the entity, see Risk governance, decision rights, and risk tolerance.

Risk treatment controls

The risk treatment step of SDAIA-P145 includes the control design and implementation task «تصميم وتنفيذ الضوابط» (p. 26). This task changes the approved treatment strategy for each risk into a treatment plan. It is possible to apply and to measure this treatment plan. The task starts with the selection of controls for the type and the source of the risk. SDAIA-P145 gives three types of controls:

• Technical controls — for example, an increase in data quality, a decrease in bias, an increase in robustness and security, and mechanisms for checks and explainability.
• Organizational controls — for example, policies, procedures, roles and responsibilities, and approval requirements.
• Operational controls — for example, procedures of use, human intervention, and controls that monitor performance and deviations.

Then each control gets a clear control owner who is responsible for the control design and implementation, and for the follow-up of the control. Each control also gets a set schedule to apply it, in proportion to the risk level and how important the risk is. The treatment plan also identifies performance indicators and success indicators that it is possible to measure. The performance indicators and success indicators show if the controls work and how much the controls decrease likelihood or impact. The performance indicators and success indicators connect to mechanisms to monitor and examine the controls continuously.

In SDAIA-P145, the treatment plan makes sure of these results (p. 26):

• The controls are applied.
• Persons can examine the controls and increase their quality continuously.
• The controls agree with the approved governance and accountability requirements.

Residual-risk reassessment

In the risk treatment step of SDAIA-P145, the residual-risk reassessment task «إعادة تقييم الخطر المتبقي» makes sure that the treatment strategies and the applied controls work (p. 27). After the controls are applied, the task estimates the risk level again. It makes sure that the risk treatment measures decreased the risks to an acceptable risk level.

The task estimates again the likelihood of the risk and its possible impact. The estimate uses the applied controls and the changes to the system design or to its operating environment. The task calculates the residual risk level with the same approved risk assessment method as before. The same method gives consistency and lets persons compare the results. For the likelihood and impact levels and the risk matrix, see Risk level and risk matrix.

The residual-risk reassessment makes sure that the risk level decreased as necessary. It also identifies one of these results (p. 27):

• The residual risk is in the approved risk tolerance limits.
• The residual risk makes more measures necessary.
• The residual risk makes escalation necessary, for a different decision.

The results of the residual-risk reassessment are recorded and connected to the risk register and to the treatment plans (p. 27). In SDAIA-P145, this record gives transparency and lets persons examine the results. It also prepares the start of the monitoring and review step during operation.

Acceptance or escalation decision

In the risk treatment step of SDAIA-P145, the acceptance or escalation decision task «اتخاذ قرار القبول أو التصعيد» makes the last decision about the AI system (p. 27). The decision is to continue the operation of the system, to release it, or to limit it. The decision uses the residual risk level and the context of use.

The task examines the results of the residual-risk reassessment. It also checks if the residual risk is in the approved risk tolerance limits. The task also uses these items (p. 27):

• How sensitive the use case is.
• The scope of impact.
• The type of the decisions that the system helps to make or makes.

Then one of these decisions follows (p. 27):

• Acceptance — the decision to accept, when the residual risk is in the acceptable risk limits.
• Escalation — for high risks, the decision goes to the applicable governance level. That level selects a conditional release, a limit on use, or a temporary or permanent stop of the system.

For this task, SDAIA-P145 recommends that entities think about these items (p. 27):

• A decision level that agrees with the risk level of the system.
• Clear records of approvals and escalation paths.

In SDAIA-P145, this record gives accountability and transparency, and lets persons examine the decisions. The conditions for the acceptance treatment option are in Treatment strategy selection. The context and scope step identifies the responsibilities of developers, operators, and decision owners, the formal approval paths, and the escalation mechanisms (see Context and scope).

Performance and risk checks

In SDAIA-P145, the monitoring and review step makes sure that controls continue to work during operation (p. 28). It also keeps the risk level of AI systems controlled during operation. The step includes continuous monitoring of performance and risks in the context of use. It monitors signs of deviation and deterioration, for example:

• Changes to the properties of the data or to the performance of the model.
• An increase in error rates or in complaints.
• Unexpected system behavior that can have an effect on fairness, safety, or compliance.

The step has three tasks: continuous performance monitoring, periodic reviews and risk register updates, and incident monitoring and response management.

In SDAIA-P145, recorded incident and periodic review results and lessons learned help continuous improvement (p. 28). Examples include an increase in control quality and changes to policies and procedures to include new information. When necessary, continuous improvement also includes a new risk assessment of the system or limits on its use.

The continuous performance monitoring task «المراقبة المستمرة للأداء» monitors the performance and the risks of the AI system during operation (p. 28). It makes sure that the system stays in the acceptable risk limits. The task identifies performance indicators and risk indicators for the context of use, for example:

• Data drift and model drift — data drift and model drift can cause a decrease in accuracy or a change in model behavior.
• Output quality — error rates, inconsistent results, and fairness indicators and bias indicators when they are available.
• Beneficiary experience — more complaints, more frequent objections to decisions, and more requests for human review.

The task also includes these items (p. 28):

• Clear alert thresholds, and response plans for when a performance indicator or risk indicator has a value above its alert threshold.
• The responsibilities for follow-up, and the reporting mechanisms.
• A connection between the results of continuous monitoring and the risk register. The results adjust the likelihood and impact estimates to new information. When necessary, the results of continuous monitoring also cause changes in risk treatment decisions.

In SDAIA-P145, continuous performance monitoring helps the entity with the early detection «الكشف المبكر» of deterioration, deviation, or unintended use (p. 28). It also makes corrective actions possible in time. Examples are a new calibration, an increase in data quality, changes to controls, or limits on use cases.

Risk register checks

The monitoring and review step of SDAIA-P145 includes the periodic reviews and risk register updates task «مراجعات دورية وتحديث سجل المخاطر» (p. 28). This task examines the risks of the AI system again on a regular basis. With this task, the estimates and the controls continue to agree with the operation of the system. This is necessary because risks change with time and when the context of use changes.

The task includes periodic reviews that follow a method, on a schedule or as event-based reviews (p. 28). Examples of changes that can start event-based reviews are:

• A change to the data.
• A change to the model.
• An increase in the scope of use.
• New regulatory requirements.

The task sets no interval for the scheduled periodic reviews.

In each periodic review, SDAIA-P145 gives these steps (p. 28):

• Estimate again the likelihood and the impact of each risk, from the new changes.
• Examine the risk treatment mechanisms and the controls, to make sure that they work and continue to be applicable.
• Use the results of applied controls and continuous monitoring to adjust the residual risk level to new information.
• Examine again the system classification by risk level or by how sensitive the system is. Make sure that it agrees with the context of use and the scope of impact at that time.

The outputs of the periodic reviews are recorded in the risk register, in a clear structure that an audit can examine (p. 28). In SDAIA-P145, these outputs contribute to consistency and let persons compare the records. The outputs also give input to risk treatment decisions, escalation decisions, and continuous improvement decisions. With these periodic reviews, AI risk management continues to change, and it does not become a procedure that stays the same.

The system classification examined in these periodic reviews is different from the risk level of each risk (see Risk level and risk matrix).

In the context and scope step, SDAIA-P145 recommends that entities think about two items (p. 17). The first item is the requirements for a new risk assessment after material changes. The other item is the mechanisms that tell users or affected groups about changes that can have an effect on performance or risk levels. See Lifecycle and change management.

Incident response

The monitoring and review step of SDAIA-P145 includes the incident monitoring and response management task «رصد الحوادث وإدارة الاستجابة» (p. 29). This task gives a method for incident response during operation, for incidents and AI system failures. The task includes these items:

• Clear reporting mechanisms for incidents or risk indicators, immediately when they are found (p. 29). This task gives no time limit as a number and names no external organization that gets the reports. The first part of the monitoring and review section also includes incident reports at the correct time, and escalation through approved governance paths (p. 28).
• Specified escalation paths that agree with the severity level and the scope of impact. These escalation paths include the competent parties and decision makers in the incident response at the correct time.
• Root-cause analysis of incidents, to identify the technical causes, the causes in how the system operates, or the organizational causes. The root-cause analysis also examines if the applied controls are sufficient.
• The necessary corrective actions and preventive actions, to make sure that the incidents do not occur again.

The incidents, the results of the root-cause analysis, and the lessons learned are recorded with a method and connected to the risk register (p. 29). These records are then used to do these tasks:

• To adjust the likelihood and impact estimates to new information.
• To increase the quality of the controls and the policies.
• To help continuous improvement and the reliability of AI systems for a long time.

Example: risk treatment

The example section of SDAIA-P145 applies the framework to a government entity that uses a large language model for internal reports (p. 32). For the context and scope of this example, see Example: internal reports. For its eight risks, see Example: risk assessment. For each risk, the example section (p. 42–44) gives a treatment strategy, implementation controls, and a decision. It also gives an estimate of residual risk after risk treatment, and measures for monitoring and review:

• Treatment strategy — mitigation for six risks. Reduction with partial avoidance for the entry or leakage of sensitive internal information. Avoidance with reduction for deliberate misuse of the system.
• Implementation controls — for example:
  • Misinformation: human review of each report is necessary before its approval. The system interface shows clearly that the outputs are not approved.
  • Sensitive information: use policies with a clear rule: entry of sensitive data or classified data is not permitted. Also, tools that prevent data leakage, and audit records of all entry and use operations.
  • Over-reliance on the system: double human review for reports that are sensitive or that have an effect on administration.
• Decision — for six risks, acceptance of the risk with continued follow-up, for example acceptance after mitigation. For sensitive information, conditional acceptance with escalation immediately when a possible incident is found. For deliberate misuse, conditional acceptance with escalation immediately on suspicion.
• Residual risk after risk treatment — the example estimates low to medium (2–4) for misinformation, and low (2) for sensitive information.
• Monitoring and review — for example:
  • Misinformation: continuous monitoring of the error rates found after use, and periodic reviews of samples of reports for accuracy.
  • Sensitive information: continuous monitoring of the use records to find when a user tries to put sensitive data into the system.
  • Over-reliance: monitor the proportion of human changes to outputs to measure reliance on the system.

For risk 7, the decrease in human competence, the example accepts the risk after the human competence measures, with continued follow-up (p. 44). It estimates that residual risk will decrease to low (2) and includes this risk in risk register checks each year (p. 44).

These measures are part of the applied example for one government entity (p. 32–44). For the measures of the example that prepare personnel, see Prepare personnel.

Related information

For the Modulos Frameworks of SDAIA-P145, their Requirements, and the steps that Modulos recommends, see Use Saudi AI Risk Management in Modulos. For the limits of Modulos risk functions, see Modulos limits. For risk identification and risk assessment, see Risk identification and risk assessment. For the FAQ about risk treatment and monitoring and review, see Saudi AI Risk Management FAQ. For other frameworks, see Comparison.

Source

• SDAIA-P145 — Saudi Data and Artificial Intelligence Authority (SDAIA), National Artificial Intelligence Risk Management Framework, version 1, April 2026 (Arabic). Risk treatment (p. 26–27), monitoring and review (p. 28–29), and the applied example (p. 32, 42–44). SDAIA.

Disclaimer

This page gives general information. It is not legal advice.