Firm AI Policy
- Safety, Security, and Robustness
- We will carefully select AI systems that meet the firm’s needs, guided by thorough research.
- Our firm will clearly define accountability for any technical errors, specifying when responsibility lies with the provider or the user.
- We will test all AI systems thoroughly before implementation and train staff on safe and acceptable AI usage in line with SRA guidance.
- Our firm will encourage staff to practice asking effective questions of AI and will provide relevant guidance to ensure optimal use.
- Transparency
- Our firm will inform clients when AI will be used in their case and explain how the AI will operate, in line with SRA requirements.
- We will follow the ICO’s guidance to ensure that clients receive appropriate explanations based on the AI’s use and its impact on them when queried.
- We will tailor our operation to ensure transparency and that we can explain how decisions are made when utilising AI.
- Fairness
- Our firm will continuously monitor AI outputs to avoid biased or inaccurate outcomes and address any bias that may emerge as AI models evolve.
- We will protect client confidentiality, especially when moving data to online systems, and ensure compliance with GDPR and SRA data protection guidance.
- We will use synthetic data to train AI systems to avoid breaches of client confidentiality.
- We will follow all data protection laws, ensuring client rights such as data removal, and will verify the jurisdiction in which any cloud-based data is held.
- Accountability and Governance
- Our firm will supervise AI systems and their usage by staff to ensure accuracy and proper function, acknowledging that AI cannot self-assess its accuracy.
- We will use AI to speed up routine tasks, always supporting rather than replacing human judgment.
- We will implement supervision processes that are capable of managing the increased speed and efficiency AI brings.
- Before deploying any AI, we will ensure a thorough understanding of the technology, and we will verify any AI-generated outputs.