Deep Learning (DL) · Advanced level

Advanced level · Module 2: Generative Models · Lesson 4 of 4

Risks and Safeguards for Generative AI

Assess harms from generated content and design safeguards for generative applications.

In this lesson you will

  • Identify harms from generative models
  • Design safeguards and disclosure practices

Key concepts

misinformationdeepfakes and consentcontent provenancehuman review

The lesson

Generated content can spread misinformation, impersonate real people, harass or deceive, and amplify bias. In fragile settings, false content about conflict or elections can cause real harm.

Safeguards include restricting uses, requiring human review before publishing, labelling generated content, refusing to generate images of real people without consent and logging use.

Teams should assess risks before deploying a generative tool and decide what it must never be used for.

Step by step

  1. List harms for a proposed generative tool.
  2. Rate likelihood and severity.
  3. Design safeguards for the top risks.
  4. Write a use policy.

Practice activity

Write the risk assessment and use policy for a generative application of your choice.

Expected result: A risk assessment with proportionate safeguards and a clear policy.

Check your understanding

  1. Why are deepfakes especially dangerous in fragile settings?

    Show answer

    False content about conflict or elections can incite harm.

  2. Name two safeguards for generative tools.

    Show answer

    For example human review before publishing and labelling generated content.

Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.