AI risk assessment for a project
Assess an AI project's risks with a clear method, a risk register and controls that are tested
A taste of a lesson
How do I handle hallucination risk in a tool that summarises customer calls?
Make it specific first. Which hallucination matters? Say: the summary invents a promise the agent never made, or omits a complaint. Score each, perhaps likelihood 3 and severity 4 for the missed complaint because of regulatory exposure. Then layer controls: preventive, constrain the summary to quote key phrases; detective, flag complaint keywords from the transcript and compare with the summary; corrective, agent confirmation before closing and weekly sampling audits. Then test them on past calls. Exercise: write one control test for the omitted complaint risk.
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Describe an AI system and its context in enough detail to assess risk
- Identify technical, human, legal and operational risks with affected stakeholders
- Score likelihood and severity consistently and map preventive, detective and corrective controls
- Test controls and assign residual risk owners and monitoring triggers
- Produce a risk register and summary suitable for sign off
Lesson plan
- 1 Describe the system Capture the system, its users and its context precisely. Start
- 2 Stakeholders and harms Identify who could be harmed and how. Start
- 3 Risk identification by category Find risks systematically across categories. Start
- 4 Scoring risk consistently Analyse likelihood and severity with defined scales. Start
- 5 Controls that are tested Map and test preventive, detective and corrective controls. Start
- 6 Residual risk, monitoring and sign off Assign owners, set triggers and document the assessment. Start
Try asking
About this tutor
For project leads, risk and compliance professionals, product owners and consultants who must assess the risks of a specific AI system before and after deployment. You follow a structured method: describe the system and its context, identify stakeholders and harms, analyse likelihood and severity, map controls, test them, assign residual risk owners and set monitoring triggers. You cover technical risks (accuracy, robustness, security, drift), human risks (automation bias, misuse, skills loss), legal and ethical risks (privacy, discrimination, transparency), and operational risks (vendor dependence, cost overruns). You work through a realistic case and produce a risk register and summary you can adapt.
Reviews
4.5
2 ratingsSample
- Amira T.Sample
Rigorous and well structured. The running case was very close to our own call centre project. Scoring scales section could include more examples.
- Gareth E.Sample
The insistence on testing controls rather than listing them changed our approach. Our register is now half the length and far more useful.
About the teacher
I help managers and teams use AI at work safely, with clear oversight and sensible rules
9 tutors 337 lessons taught Sample
I teach AI for the workplace from the angle of people who have to make decisions and answer for them. My learners are managers, team leads and staff who want to use AI without leaking data, misleading customers or handing judgment to a machine. We cover safe everyday use, oversight that works, regulation in broad terms, data protection principles and...
See Alejandra's profile and tutorsMore like this
Other tutors on the same or nearby topics.