AI Disclaimer

Nature of AI Outputs

AI systems — including those based on large language models (LLMs), machine learning, and other AI techniques — do not "know" information in the way humans do. They generate outputs based on patterns in training data and contextual inputs.

As a result, AI outputs may:

• Contain factual errors or inaccuracies, including confident-sounding but incorrect statements (sometimes called "hallucinations")
• Be incomplete, outdated, or not reflect current information
• Produce different outputs for the same or similar inputs on different occasions
• Reflect biases present in training data
• Fail to account for context, nuance, or edge cases
• Degrade in performance over time as the underlying world changes

No AI system built by AI by EDM should be treated as a definitive, error-free source of information or decision-making without appropriate human review.

No Professional Advice

AI systems we build do not constitute and must not be used as a substitute for professional advice in any regulated domain, including but not limited to:

• Medical and clinical advice — AI systems must not replace clinical judgement or diagnosis by a qualified medical professional
• Legal advice — AI outputs are not legal advice and must not be relied upon as such without review by a qualified solicitor or barrister
• Financial advice — AI outputs do not constitute regulated financial advice under the Financial Services and Markets Act 2000
• Engineering or safety-critical advice — AI outputs must not be used to make safety-critical decisions without expert human review

Where AI systems are built to assist in regulated domains (such as the PLAB 2 medical education agent), they are designed as decision-support and learning tools only, not as replacements for qualified professional judgement.

Human Oversight Requirement

Clients who deploy AI systems built by AI by EDM are required to maintain appropriate human oversight of those systems. This means:

• A qualified person reviews AI outputs before they are acted upon in high-stakes contexts
• Automated AI decisions that affect individuals are subject to a human review mechanism
• There is a clear process for identifying, reporting, and addressing AI errors or unexpected behaviour
• Staff who interact with or rely on AI outputs are trained to understand the system's limitations

This requirement is particularly important for AI systems operating in healthcare, finance, legal, HR, or any other regulated or sensitive domain.

Data and Bias

AI systems learn from data. The quality, diversity, and representativeness of that data directly affects the quality and fairness of AI outputs.

• AI systems may produce outputs that reflect historical biases present in training data
• AI systems trained on limited datasets may not perform equally well across all user groups or contexts
• We apply reasonable steps to identify and mitigate bias during system development, but cannot guarantee that bias has been fully eliminated

Clients are responsible for monitoring deployed AI systems for bias or discriminatory outcomes and taking corrective action where identified. AI systems must not be used to make decisions that unlawfully discriminate against individuals under the Equality Act 2010 or applicable law.

Third-Party AI Models

Many AI systems we build incorporate foundation models from third-party providers such as OpenAI, Anthropic, Google, and others. We are not responsible for:

• Changes, updates, or deprecation of third-party models that affect system performance
• Outages or availability issues with third-party AI model APIs
• Changes to third-party provider usage policies that may affect permitted use cases
• Outputs generated by third-party foundation models that fall outside our control

Clients should familiarise themselves with the terms of service of any third-party AI models incorporated into systems built for them.

Security Limitations

AI systems may be subject to adversarial attacks including prompt injection, jailbreaking, data poisoning, and other techniques designed to manipulate AI behaviour. We implement reasonable security measures during system design, but:

• No AI system is completely immune to adversarial manipulation
• Clients are responsible for implementing appropriate access controls on deployed systems
• Unusual or unexpected AI behaviour should be reported to us promptly

Liability

AI by EDM shall not be liable for any loss, damage, or harm arising from:

• Reliance on AI outputs without appropriate human review
• Use of AI systems in a manner inconsistent with the intended purpose or in violation of these terms
• Decisions made based on AI outputs in regulated or high-stakes domains without qualified professional oversight
• Changes to third-party AI model providers that affect system performance
• Adversarial attacks or misuse of AI systems

Our liability is further governed by the Terms & Conditions.

Regulatory Compliance

The regulatory landscape for AI is evolving rapidly in the UK and internationally. Clients are responsible for:

• Understanding the regulatory requirements that apply to their specific use of AI
• Ensuring AI systems are deployed in compliance with the UK GDPR, the Data Protection Act 2018, and any sector-specific regulations
• Monitoring developments in UK AI regulation and updating their approach accordingly

We will use reasonable endeavours to build systems that support regulatory compliance, but we cannot guarantee compliance with regulations that may change after system delivery.

Questions