MIT Technology Review reports caste bias in OpenAI products

MIT Technology Review reports caste bias in OpenAI products

Published October 1, 2025, MIT Technology Review’s The Download says its investigation found caste bias in ChatGPT and Sora, OpenAI’s text-to-video generator. It argues that these products reproduce stereotypes portraying Dalits as dirty, poor, and suited only to menial jobs, despite Dalit people’s achievements in fields including medicine, public service, and scholarship. The newsletter warns that this bias risks entrenching discriminatory views without being addressed. Its concern is not only how AI portrays people, but how those portrayals may sustain stereotypes that conflict with the lives and work of caste-oppressed people. The retrieved newsletter does not explain how the investigation assessed the products or what steps might address the reported bias, so the extent and possible remedies remain unclear.

Publication: 87943fce-911c-e551-be0a-0484db66e657

attribution — pass: The page is one newsletter issue in which MIT Technology Review comments in its own words, so it counts as a single piece. The reading correctly attributes the caste-bias finding to MIT Technology Review's investigation, as summarized in The Download, and keeps it as the publication's claim rather than established fact.

Caste bias is rampant in OpenAI’s products, including ChatGPT, according to an MIT Technology Review investigation.

summary — pass: The reading clearly states the actor, the products and the stereotypes, and it labels Dalits as caste-oppressed people. One minor drift: the source says models render Dalits as 'performing only menial jobs', while the reading says 'suited only to menial jobs'. Using 'performing' would be closer, but the meaning of a stereotype is preserved. The title and question follow from the evidence.

render Dalits as dirty, poor, and performing only menial jobs

date — pass: October 1, 2025 matches the UTC calendar day of originalPublished and the URL. The kind is publisher-page, so no archive-copy disclosure is required. Later-dated Deep Dive sidebar items, such as the 377 math findings, are possible later additions; the reading neither uses nor relies on them. No hindsight framing appears.

support — pass: Both claims are verbatim-supported. The concern about entrenching discriminatory views is attributed to the source. The limitation noting that methods are absent is accurate. The question about remedies follows from the gap the reading identifies and is not leading.

This risks entrenching discriminatory views in ways that are currently going unaddressed.

media — pass: No media was selected, and no verified candidates were supplied. Nothing was invented.