Tag: AI TRiSM

By Anahita Bilimoria, Decision Lab Innovation Practice Lead

Welcome back to our series on AI TRiSM! In our previous post, we established that Trust is the necessary foundation for AI adoption, built on principles of explainability, fairness, and reliability. However, even the most trusted system carries inherent uncertainties.

The Illusion of Certainty

It is a fundamental fact of data science: while every system is modelled on reality, no model can be a perfect reflection of the real world. Even outside of AI, we accept risk in our most trusted systems:

• Climate Change Models: These are trusted for predicting future warming, yet they involve significant uncertainty (offering a range of possible outcomes) due to the necessary simplification of complex atmospheric, oceanic, and biological interactions.
• Cybersecurity: Highly trusted software systems are constantly patched because determined attackers find zero-day vulnerabilities—flaws the designers didn’t know existed.
• Aviation: While pilots and air traffic control are highly trained, risk is always present due to potential miscommunication or procedural lapses. Checklists and redundancy are built-in specifically to manage this uncertainty.

It is safe to assume that risk is inherently present in all solutions. This brings us to the second, equally crucial pillar of the AI TRiSM framework: Risk Management.

Responsible AI deployment is not about eliminating risk entirely—that is impossible. It is about establishing an effective, proactive strategy for identifying, quantifying, and mitigating it.

In this post, we will:

1. Distinguish between traditional IT risk and unique AI risk.
2. Categorise the specific harm vectors relevant to AI.
3. Outline a four-step framework to operationalise risk management in your organisation.

AI Risk is Not Traditional IT Risk

In traditional IT and cybersecurity, risk is primarily focused on system availability, data security, and compliance breaches. While these still apply, AI introduces unique vectors of harm that require a specialised approach.

The challenge is that AI risks are often non-deterministic—they are tied to the model’s behaviour, not just the infrastructure.

Because these risks move beyond simple system failure, they are trickier to quantify and mitigate.

Categorising AI Risk: The Harm Vectors

To manage AI risk effectively, organisations must classify potential harms into structured categories. These categories provide a blueprint for assessment along with standard mitigation strategies.

1. Performance and Operational Risk

This refers to the risk of the model failing to deliver its intended technical outcomes, or its performance degrading in a real-world environment. This directly impacts Cognitive Trust.

• Model Drift: The model’s real-world data distribution shifts away from the training data, causing accuracy to drop.
  • Mitigation: Implement robust ModelOps monitoring pipelines that continuously compare production performance against established baseline metrics. It is imperative to create pipelines that detect data drift above a certain threshold. If significant drift occurs, the model can be retrained on new data to restore accuracy—frameworks like AgileRL can be instrumental here, offering efficient evolutionary algorithms to accelerate these retraining cycles.
• Adversarial Attacks: Malicious actors introduce subtle, often imperceptible, changes to inputs that trick the model into misclassification (e.g., making a stop sign look like a yield sign to a self-driving car).
  • Mitigation: Employ Adversarial Training during development. Furthermore, organisations can mitigate the risk of non-deterministic AI outputs by pairing them with deterministic Mathematical Optimisation (such as Gurobi). This ensures that even if an AI model acts unpredictably, the final decision is bounded by hard constraints that prevent unsafe or illogical actions.

2. Ethical, Societal, and Reputational Risk

These are risks related to unfairness, bias, lack of transparency, or the unintended negative impact of the AI system on individuals or society. This directly impacts Emotional Trust and brand integrity.

• Bias and Discrimination: The system perpetuates or amplifies historical biases, leading to unfair decisions in high-stakes contexts (e.g., loan applications, hiring, or criminal justice).
  • Mitigation: Conduct Fairness Audits using techniques like disparate impact analysis across protected groups. Implement bias mitigation techniques at every stage of the solution lifecycle. Exploratory Data Analysis (EDA) should be used to highlight data skew that could lead to a biased model.
• Lack of Explainability: The black box nature prevents users or regulators from understanding why a decision was made.
  • Mitigation: Prioritise XAI (Explainable AI) techniques like SHAP and LIME for black-box models, especially in high-consequence decision-making. Where possible, employ inherently white box models (such as Logical Neural Networks) for inbuilt transparency.

3. Security and Compliance Risk

This covers risks related to data privacy, intellectual property theft (model inversion/extraction), and regulatory non-compliance.

• Data Leakage/Privacy Violation: The model inadvertently reveals sensitive training data during inference.
  • Mitigation: Employ Federated Learning (FL), where the model is trained on decentralised edge devices (like smartphones) or local servers. Only model updates (gradients)—not raw data—are sent to the central server. Additionally, Data Sanitisation and Anonymisation ensure that Personal Identifiable Information (PII) is stripped, preventing data from being linked to individuals.
• Regulatory Fines: Failure to adhere to region-specific AI regulations (e.g., the EU AI Act).
  • Mitigation: Establish an AI Governance practice responsible for classifying systems by risk tier. Platforms like Red Hat OpenShift AI can automate this governance, ensuring that mandatory documentation, security protocols, and testing requirements are enforced as a standard part of the solution lifecycle.

Operationalising Risk Management: The Assessment Framework

A responsible organisation integrates AI risk assessment into its existing Enterprise Risk Management (ERM) framework. This process involves four steps:

1. Risk Identification: Map the AI system’s use case to potential harm vectors (e.g., A loan approval model has a high bias risk or A real-time recommendation engine has high model drift risk).
2. Risk Quantification: Estimate the likelihood of the harm occurring and the potential impact (financial, reputational, or societal severity). To do this effectively, organisations can use simulation technology—specifically Digital Twins built with tools like AnyLogic—to test AI models in a risk-free virtual environment before real-world deployment.
3. Risk Mitigation: Implement controls (as listed above) to reduce likelihood and/or impact.
   • Note on Insurance: While software liability is standard, the industry is increasingly discussing AI-specific liability insurance. This emerging sector aims to cover the unique, non-deterministic risks of AI agents that traditional policies might miss.
4. Risk Monitoring: Establish continuous monitoring mechanisms (the Monitoring pillar of TRiSM) to ensure controls remain effective and to catch emerging risks quickly.

The Mandate of Proactive Risk Management

The era of merely deploying a high-performing model and hoping for the best is over. Regulatory bodies across the globe are increasingly making proactive risk assessment a legal mandate.

The AI TRiSM framework provides the discipline to make this transition. It shifts the focus from simply maximising performance metrics to optimising for outcomes across performance, ethics, and security.

By adopting a structured approach to risk, organisations don’t just protect their bottom line—they solidify the trust built with their users and ensure their AI systems are safe, ethical, and sustainable for the long term.

Contact Decision Lab today to learn how our TRiSM-aligned strategies can secure your AI initiatives.

By Anahita Bilimoria, Decision Lab Innovation Practice Lead

An essential framework for Responsible AI Deployment

In this blog post, we dive deeper into the AI TRiSM principle of Trust. Together, the principles of AI TRiSM (Trust, Risk, and Security Management) add transparency, understandability, and reliability to our AI systems.

Continuing from our previous blog on AI TRiSM, Building Trust in the Age of Artificial Intelligence, where we took a holistic view of the three core pillars of AI TRiSM, this blog post dives deeper into the principle of Trust. Welcome to part two of our AI TRiSM series: Trust in AI systems.

The Foundation of Adoption: Building Trust in AI

AI has integrated into our daily lives in unprecedented ways, from using Gemini or ChatGPT to summarise reports to utilising tools like Google’s NotebookLM for learning. However, the reliability of answers given by AI systems often remains uncertain. While we confidently ask a Language Model for code, broad trust in AI systems is still a major concern.

Trust: An assured reliance on the character, ability, strength, or truth of someone or something

Webster’s dictionary

In the context of AI, a more fitting definition may be: the attitude that an AI agent will help achieve an individual’s goals in a situation characterised by uncertainty and vulnerability. Clearly, building trust goes beyond simple belief in their capabilities—AI chatbots can seem remarkably confident. Rather, we must establish a calibrated trust: confidence that an AI system will behave reliably, ethically, and securely, leading to intended outcomes while understanding its limitations. For the human in the system, this confidence operates on two crucial dimensions:

• Cognitive Trust: based on evidence, competence, and reliability. Confirmed by performance metrics such as accuracy, loss, and F1 score, ensuring the system works as expected. Answers the question: can I trust it?
• Emotional Trust: based on comfort, security, and ethical alignment. The confidence that the system aligns with moral or societal values and will not discriminate against users. Answers the question: do I want to trust it?

The goal of AI TRiSM is to foster a balanced level of trust across these dimensions, steering clear of algorithm aversion (distrusting a competent system) and automation complacency (over-relying on a flawed system). Trust in AI is not a single feature but the culmination of several measurable and governable qualities. The AI TRiSM framework tackles these factors by providing actionable strategies:

1. Explainable and Transparent (XAI)

The defining challenge for trust is the ‘Black Box’ nature of modern, complex AI models. If a system’s decision cannot be understood or audited, it fundamentally cannot be trusted.

Explainable AI (XAI) addresses this by providing insight into how and why a model reached a specific output. This is essential not just for a user’s peace of mind, but for auditing, compliance, and legal accountability.

How it can be achieved:

• Employing Inherently Interpretable Models: Using simpler models (like linear regression or decision trees) when complexity isn’t strictly necessary. Utilising inherent white-box models such as NeSy models (Neurosymbolic AI) and Causal ML routes that provide complete knowledge graphs of model knowledge.
• Providing Decision Tracing: Logging all input features and intermediate steps that lead to an outcome, allowing stakeholders to trace a decision back to its source. Applying methods like SHAP (SHapley Additive exPlanations) or LIME (Local Interpretable Model-agnostic Explanations) to complex models to generate human-readable justifications for individual predictions.

2. Fairness and Bias Mitigation

AI models are trained on data, and that data reflects historical and societal biases. Consequently, models are prone to inheriting these biases, leading to discriminatory or unfair outcomes. This directly breaks emotional trust. Building trust requires active and continuous steps to ensure fairness.

How it can be achieved:

• Pre-Processing Bias Mitigation: Conducting thorough Exploratory Data Analysis (EDA) to identify and balance data imbalances before model training (e.g., re-sampling minority classes).
• Model Bias Mitigation: Implementing constraints during the training process that penalize the model for differential performance across different demographic groups. Defining and monitoring multiple fairness metrics (like equal opportunity or demographic parity) after deployment to ensure equal outcomes across protected groups, going beyond simple overall accuracy.

3. Reliability, Robustness, and Safety

A trusted system must be dependable. Reliability is its ability to perform consistently and accurately under normal operating conditions. Robustness ensures the model’s accuracy is maintained even when facing slight variations or unexpected inputs. The final layer is Safety, which protects against catastrophic failure.

How it can be achieved:

• Continuous Model Operations (ModelOps): Implementing automated systems to monitor model performance in real-time, catching model drift (when performance degrades over time) or degradation after deployment.
• Stress Testing and Adversarial Training: Rigorously testing the model with malicious inputs and unexpected data shifts to improve robustness against adversarial attacks.
• Human-in-the-Loop Controls: Equipped with safeguards like “kill switches” and defined pathways for human intervention to ensure an autonomous system can be overridden or stopped when faced with an unsafe or ambiguous situation.

4. Privacy and Data Protection

In the age of vast data collection, a user’s willingness to use an AI solution hinges on the assurance that their sensitive information will be protected. Trust is lost if data is compromised, misused, or leaked. Adhering to regulations like GDPR and CCPA is a baseline.

How it can be achieved:

• Secure-by-Design Principles: Initiating and maintaining AI solution development that incorporates techniques such as anonymisation and data minimisation (minimum necessary data collection).
• Privacy-Enhancing Technologies (PETs): Utilising advanced cryptographic techniques like federated learning (training models on decentralised data) to protect sensitive information during training and inference.
• Access Control and Security Audits: Implementing strict access controls and regular security audits for data pipelines and model APIs to ensure compliance and prevent unauthorised data access.

The Value of Proactive Trust Management

The journey of AI adoption is paved with the potential for misuse and technical failure. A single, high-profile failure, such as a biased recommendation, a security breach, or a dangerous hallucination, can instantly erode years of trust-building effort.

By embracing the Trust component of AI TRiSM, organisations can move from reactive damage control to proactive trust management. They can operationalise these ethical and performance requirements, embedding them into the entire solution lifecycle.

Investing in these principles is an investment in the long-term viability of AI, ensuring that as systems become more autonomous and integrated into our lives, they remain aligned with our values, transparent in their operations, and secure with our data. This is how we build systems that don’t just perform a task reliably, but adapt and improve in a volatile world. This concept is particularly critical in complex, high-stakes environments like supply chain management, where disruption is a constant threat.

For a deeper exploration of how these principles are applied to create systems that gain from disorder, see our recent white paper: Beyond Resilience: Engineering the Anti-Fragile Pharma Supply Chain of 2030.

Author: Anahita Bilimoria, Decision Lab Innovation Practice Lead.
Follow this series on AI TRiSM from Decision Lab, follow us on LinkedIn!

By Anahita Bilimoria, Decision Lab Senior Machine-Learning Engineer

An Essential Framework for Responsible AI Deployment

The promise of Artificial Intelligence is immense, offering solutions to humanity’s most pressing challenges. With the recent boom of Language Models and AI penetrating every domain, we are confronted with a fundamental truth: global adoption and subsequent progress rely solely on trust. While the performance of AI in every aspect of automated decision-making is phenomenal, there has been a rising concern for trust in AI, fueled by opaque decision-making and perceived biases. This challenge has given rise to stalled innovation, public apprehension and the risk of deploying technologies without adequate oversight. The urgency to build and maintain trust in AI has surpassed being simply a matter of ethics and has given rise to safety concerns as well.

Despite adaptive speed and response, autonomous AI’s unchecked deployment can lead to instability. The absence of clear dependability metrics and inadequate interpretability methods raise trust questions across diverse AI. Although some transparent AI exists, rapid critical adoption in a volatile environment, coupled with regulation and public concern, necessitates urgent trust-building. AI TRiSM provides a repeatable framework (trust, security, privacy, transparency) to address these risks.

What is AI TRiSM?

Gartner, a leading research and advisory company, defines AI TRiSM (AI Trust, Risk and Security Management) as a framework that ‘ensures AI model governance, trustworthiness, fairness, reliability, robustness, efficacy, and data protection’. This ensures that models that follow this framework are not unethical, unfair or biased. While most AI solutions focus on model performance, AI TRiSM adds a layer of model responsibility, urging developers to strike a balance between the two.

Figure 1: Balancing Model Performance and Model Responsibility. This diagram illustrates the key metrics associated with traditional AI Model Performance (such as Accuracy, F1 score, and Loss) in contrast with the crucial metrics for Model Responsibility (including Transparency, Explainability, Fairness, and Security) that are central to AI TRiSM.

Despite being a framework of individual principles, AI TRiSM enables the fulfillment of each principle through conscious and targeted steps.

Trust

Entailing the concepts of transparency, fairness, reliability, privacy and safety; this pillar ensures that models offer accountability and build trust. This component of AI TRiSM requires models to offer explainability, either by using Explainable AI (xAI), which are models designed to be interpretable, or artificially inducing explainability in their decision-making processes. Models trained on data are prone to biases in the data itself, leading to discriminatory model outcomes. Techniques like a thorough Exploratory Data Analysis (EDA) – which involves visualizing and summarizing data to spot imbalances – and bias detection methods can help gain insight into the biases in data. Regulations like GDPR and CCPA ensure data privacy and security. While models can’t follow these regulations directly, you can ensure your AI solution does by implementing appropriate data handling and storage practices! Autonomous solutions can build trust by introducing kill switches, pathways for human intervention, and safe operation mechanisms to ensure safe execution even in unexpected situations.

Risk

This pillar involves identifying and managing risks associated with your AI solution throughout its lifecycle. You can map up the risks associated with your solution. Some key aspects can include:

• Performance risks (e.g., model drift, accuracy degradation)
• Ethical risks (e.g., bias, lack of fairness)
• Security risks (e.g., adversarial attacks, data breaches)
• Operational risks (e.g., deployment failures, integration issues)

The solution lifecycle must include allocated resources for identifying, evaluating, and mitigating risks proactively throughout the solution’s development and deployment. Identifying these risks proactively helps stakeholders make informed decisions about the deployment and ongoing use of the AI solution. Ensuring the risk register includes all potential risks will highlight the potential gaps in your solution and enable mitigation strategies. Your team can have established roles, responsibilities, and policies for managing AI risks effectively.

Security

Each type of AI comes with its own security issues, like adversarial attacks (subtly changing input data to fool the AI), data poisoning (injecting malicious data to corrupt training), and model stealing (recreating a proprietary model). Achieving security in the model performance involves achieving security throughout the solution lifecycle, following a ‘Security by design’ approach to developing your solution. One must secure their data (incoming and outgoing), adopt techniques to protect the AI models, infrastructure, and APIs. While this principle shares its goal with standard security for software solutions, in AI TRiSM this also means making sure your model is not ‘hackable’ in ways specific to AI vulnerabilities.

How can you adopt AI TRiSM as a company?

While developers can tackle components individually, companies can also embrace it as a complete framework. The following steps outline how your company can introduce AI TRiSM to your organisation:

• Adopt AI TRiSM across your entire solution lifecycle (discovery to evaluation), documenting observations and decisions in a final report. A company-wide template standardizes AI TRiSM implementation for all projects.
• Ensure company-wide awareness of AI TRiSM through clear communication channels. Document model audits and communicate identified risks with proposed mitigation strategies to relevant stakeholders.
• Provide thorough training and education on AI TRiSM principles and practices across the organization, perhaps through workshops or online modules.
• Establish partnerships with entities that have a strong focus on AI TRiSM to leverage their expertise and insights.

Employing TRiSM in AI offers several key benefits, including building trust in AI systems, mitigating potential risks through pre-emptive resolution, ensuring compliance with evolving regulations, and fostering sustainable and transparent AI growth.

This post has provided an overview of AI TRiSM and its critical role in the responsible development and deployment of AI. In upcoming articles, we will delve deeper into each of the core pillars – Trust, Risk, Security, and Transparency – exploring the specific challenges, techniques, and best practices associated with building trustworthy AI systems.

Investing in AI TRiSM is an investment in the long-term value and viability of AI. By embedding these principles into our processes, we build a foundation of trust that will be crucial for the continued adoption and positive impact of artificial intelligence!

Next Steps

Navigating the complexities of AI TRiSM – ensuring trust, managing risk, and maintaining security – is crucial for successful AI adoption. At Decision Lab, we are committed to developing AI solutions that inherently incorporate these principles from design to deployment.

Our deep expertise in Explainable AI (xAI) provides the transparency needed for user confidence and regulatory compliance, directly addressing the ‘Trust’ pillar. We specialise in creating effective Human-AI Teaming paradigms, designing systems where human insight complements automated decision-making, ensuring robust operation and essential oversight to mitigate ‘Risk’.

Furthermore, our development processes are underpinned by strict adherence to rigorous ISO standards, demonstrating our commitment to ‘Security’, reliability, and quality across all our AI solutions. Partner with Decision Lab to build AI systems that are not only high-performing but also fundamentally trustworthy, secure, and aligned with responsible innovation principles.

To explore how Decision Lab’s AI solutions can benefit your organisation, get in touch. Let’s unlock the full potential of AI.

The next post in the series Trust in AI Systems: An essential framework for Responsible AI Deployment.

Author: Anahita Bilimoria, Decision Lab Senior Machine Learning Engineer
For further updates from Decision Lab, follow us on LinkedIn!