1. EXECUTIVE SUMMARY 7

2. TECHNICAL BACKGROUND TO GENERATIVE AI 11

2.1 LARGE LANGUAGE MODELS (LLMs) 12

2.2 GENERATIVE ADVERSARIAL MODELS (GANs) 12

2.3 VARIATIONAL AUTOENCODERS (VAEs) 12

2.4 DIFFUSION MODELS 13

2.5 STANDALONE GENERATIVE AI TOOLS 13

3. RELEVANT LEGAL STANDARDS 14

3.1 THE RIGHT TO PRIVACY 14

3.2 THE RIGHT TO EQUALITY AND NON-DISCRIMINATION 15

3.3 THE RIGHT TO FREEDOM OF EXPRESSION 17

3.4 FREEDOM OF THOUGHT 18

3.5 BUSINESS AND HUMAN RIGHTS STANDARDS 19

4. BACKGROUND TO GLOBAL DEVELOPMENTS IN GENERATIVE AI 21

5. HUMAN RIGHTS TENSIONS IN GENERATIVE AI DESIGN 25

5.1 UNLAWFUL WEB-SCRAPING AND MASS INVASIONS OF PRIVACY BY DESIGN 25

5.2 DATA CENTRES AND ENVIRONMENTAL IMPACT 28

5.3 PERPETUATING STEREOTYPES, BIASES AND DISCRIMINATION 30

5.4 SHRINKING CIVIC SPACE 32

5.5 AUTOMATION BIAS AND MANIPULATION 34

5.6 CONCLUSION 35

6. HUMAN RIGHTS INCOMPATIBILITY OF CURRENT GENERATIVE AI SYSTEMS BASED ON DATA PIPELINE 37

6.1 THE RIGHT TO PRIVACY 37

6.2 THE RIGHT TO EQUALITY AND NON-DISCRIMINATION 38

6.3 RIGHT TO FREEDOM OF EXPRESSION 38

6.4 RIGHT TO FREEDOM OF THOUGHT 39

6.5 BUSINESSES’ HUMAN RIGHTS RESPONSIBILITIES 39

7. CONCLUSIONS AND RECOMMENDATIONS 40

7.1 CONCLUSION 40

7.2 RECOMMENDATIONS 40

TO STATES 40

TO COMPANIES 41

TO UN BODIES AND REGIONAL ACTORS 41