This paper shows that the equity market prices the novel social risks associated with generative AI. We exploit the release of ChatGPT as an information shock that updated investor beliefs about AI-related tail risks. Using pre-event ESG scores as proxies for firms' social risk management, we show that, controlling for AI productivity exposure, low-ESG firms underperform high-ESG firms by 4 percentage points over the two weeks following the release. The effect is concentrated in the Social pillar, particularly data privacy and security. Increases in option-implied downside risk indicate that changes in discount rates are the channel. A low-minus-high ESG portfolio of AI-exposed firms earns significant alphas during 2023–2024, suggesting that investors demand compensation for bearing AI-related downside risks. Our findings are consistent with a tail-risk model in which social risks of AI are priced through discount rates.