Technology and Digital Systems
The Global Centre for Risk and Innovation (GCRI)
Risk Academy, a Nexortium programmeTechnology and Digital SystemsDepartment of Artificial Intelligence and Autonomous Systems
Evaluation design, red-teaming, benchmarks and safety cases
AI safety and evaluation covers how the capabilities, reliability and risks of AI systems are tested before and after deployment.
Back to Department of Artificial Intelligence and Autonomous Systems
risksacademy.comtechnology-digitaldiscipline
Twin figures for West Cornwall: Goonhilly and the Porthcurno cable landings, counted from its layers at their stated dates.
For AI safety and evaluation: the faculty's twin of the Lizard and Penwith on the planet map, the globe and in 3D. Pick a generator, switch layers, run the instruments; every layer prints its provider and date.
Drag to move, right drag or Shift with a drag to turn and tilt, Ctrl with the wheel to zoom. Keyboard: arrows move, Shift with arrows turn and tilt, + and - zoom, 0 resets, V steps through the views, Escape clears.
Flat plan of the same data: relief as contours from the inline grid, the mapped ground infrastructure by kind, the mapped cables and the numbers below. The 3D twin opens on devices with WebGL once its data and library have been checked.
The map loads from this site as it nears the view (MapLibre 6.9.1, checked by hash).
The discipline teaches evaluation design with stated tasks, metrics and evidence, red-teaming and adversarial testing, benchmark contamination and evaluation awareness, dangerous capability evaluations, monitoring after deployment and incident reporting, and the safety case as the argument that ties them together, following the holistic evaluation taxonomy of Liang et al. (2022) for metrics and the NIST AI Risk Management Framework 1.0 for the measure function, with the International AI Safety Report of February 2026 as the current evidence review. The evaluation ledger on the faculty home is the discipline's working instrument: records with evidence links, summary statistics per metric, unsupported entries marked. A graduate can design an evaluation that a benchmark alone cannot answer, run a red-team exercise with a stated scope, and write a safety case that names the hazards it does not cover.
This area covers how the capabilities, reliability and risks of AI systems are tested before and after deployment: evaluation design, red-teaming, benchmarks and their limits, safety cases for frontier models and agentic systems, and incident reporting. It is taught in the department of Artificial Intelligence and Autonomous Systems, Faculty of Technology and Digital Systems, through courses still to be authored.
8 subtopics, taught across its courses and labs.
Not yet taught by a course: AI system evaluation methods; Frontier AI risk and safety cases.
Recent developments in aI safety and evaluation, each with its source; sources checked on 16 October 2026. Summaries are short readings of the source; follow the link for the full account.
News and research as they are published, per discipline, with the faculty's other tools.
The newest research, preprints, news and humanitarian reports for AI safety and evaluation, each with its source and date, load here. Open the research and news hub
No course has this as its primary discipline yet. It is listed so the gap stays visible.
The courses on its authoring list are named below.
Approved for authoringno enrolment and no achievement until authored and reviewed
One for each of the 0 courses with this primary discipline, when you pass its assessed activities; 0 can be earned now.
Achievements record learning completed here; they do not certify competence. How recognition works.
Cited as places where the subject is practised; the institutions named are sources here, and Risk Academy hosts every course.
Open courses in connected disciplines teach the methods this discipline will build on.