2.0 Legally compliant

#

 Principle Objective Guideline
2.1 Risk and Impact Assessed We will conduct assessments of the social, economic, and environmental risks and impacts of AI in our products and services

Identify the risk level of any potential functionality, aligned with the capAI system

Ensure AI-specific risk assessments are completed, considering key elements like privacy, safety etc

Address potential negative impacts promptly
2.2 Security We commit to implementing best-in-class security measures to safeguard AI systems and the data they process

Leverage encryption and access controls wherever appropriate

Conduct regular security audits and penetration tests
2.3 Compliance & Governance We will adhere to all relevant laws and regulations governing the use of AI, including proactively implementing guidance and processes for legislation under development

Adhere to all relevant laws and regulations governing the use of AI, including data protection, privacy, and anti-discrimination laws

Enforce data governance practices that adhere to data protection regulation

Keep abreast of any upcoming or evolving legislation

Ensure any third-party tools or services that leverage AI have been risk assessed

 

3.0 Technically robust

#

 Principle Objective Guideline
3.1  Interpretability & Transparency We’ll aim to ensure that stakeholders can interpret and explain the rationale behind the outputs of our AI

Produce thorough, user-friendly documentation for every project or process that leverages AI
3.2  Lifecycle Quality Assurance We will ensure quality at each of the key AI system lifecycle stages, from design through to development, evaluation, operation and if applicable, retirement

Implement knowledge transfer processes to ensure insights gained from ongoing quality assurance efforts are shared across Product & Technology teams
3.3 Testing & Monitoring We will take measures to enhance the reliability, accuracy, and performance of our AI systems and monitor against these our benchmarks

Test under a range of conditions, including diverse environments, input variations, and different user demographics

Ensure systems maintain reliability and performance across a spectrum of scenarios

Set up mechanisms for continuous monitoring of performance in real-world conditions

Implement alert systems to detect deviations from expected behaviour
3.4 Sustainability & Scalability We will seek to develop AI products and systems that have longevity and can scale aligned to the goals of our users

Take an agile and progressive approach to development and deployment

Where possible build reusable capabilities rather than tying delivery tightly to a narrow use-case

Consider adopting third party tooling where it allows for, experimentation and changes to implementation, whilst adhering to our AI policy
3.5 Automated Evaluation Measure the accuracy and quality of AI systems

Use Evals to automatically test before every deployment
3.6 Maintainability Maintain necessary expertise and resource within the business to support AI projects post-launch

Automatic tooling for security updates, which re-run tests

 

ICON-contract

AI policy template for institutions

We've created a comprehensive AI Policy template, readily available for institutions to download and customize, providing a robust framework for responsible AI implementation.
Download the template