AI-Assisted Engineering
Accountability and Equal Standards
Summary
The directing engineer remains fully accountable for AI-assisted work, which meets the same principles and standards as work authored entirely by a human.
Reasoning
An AI tool does not assume responsibility for the correctness, quality, security, or compliance of its output. The engineer directing its use remains responsible for evaluating and maintaining every resulting engineering artifact.
The method used to produce engineering work does not change the outcomes expected of it. AI assistance does not change which principles and standards apply.
Implemented By These Standards
Risk-Proportionate Scrutiny
Summary
Scrutiny of AI-assisted work increases with its security, safety, or regulatory impact and includes relevant domain expertise for high-risk work.
Reasoning
The consequence of an incorrect output differs by context. A defect in authentication, clinical logic, or another high-impact capability can cause harm beyond the code path in which it appears.
Additional review by people who understand the affected domain provides context that general code review cannot supply and keeps the depth of verification proportionate to the potential impact.
Implemented By These Standards
Understanding Before Adoption
Summary
AI-assisted output is adopted only once the responsible engineer understands it well enough to explain, justify, and maintain it.
Reasoning
Without understanding an output's behaviour and implementation, the responsible engineer cannot explain its decisions or maintain the result. Faster production does not replace that understanding.
Implemented By These Standards
AI Contribution Traceability
Summary
A material AI contribution remains disclosed with enough context to identify its role in a code change or incident response.
Reasoning
Accountability depends on knowing where AI assistance materially influenced an engineering outcome. Disclosure gives reviewers the context needed to assess an assisted change and preserves an accurate record of how incident conclusions and actions were reached.
Identifying the affected work is more useful than recording AI use without its scope. It allows later investigation to distinguish an AI contribution from the human decisions that accepted or acted on it.
Implemented By These Standards
Suitable AI Tools
Summary
Engineering work uses AI tools whose terms and controls are suitable for the work and information entrusted to them.
Reasoning
AI tools differ in how they protect and use organisational code and data. Suitability depends on whether a tool's terms and controls are acceptable for the work and information involved. Capability, convenience, and cost do not establish suitability on their own.
Implemented By These Standards
Intellectual Property
Summary
AI-assisted output is adopted only when any third-party material it contains can be used lawfully.
Reasoning
AI-generated output can contain third-party material without carrying the right to use it. Reviewing output before adoption prevents that material from entering an engineering artifact without permission.
Implemented By These Standards
Confidentiality
Summary
Confidential material is submitted to an AI tool only when its handling terms preserve that confidentiality.
Reasoning
Submitting confidential or proprietary material to an AI tool can expose it to provider retention, reuse, or access. Restricting submissions to tools whose handling terms preserve confidentiality keeps the material within its intended controls.