AI incident register
A public record of cases in India where an AI system harmed people, failed in a way that mattered, or was used to deceive them. Each record says what happened, what kind of system was involved and what followed, with links to its sources.
- Finance and payments: 4
- Courts, tribunals and policing: 4
- Online platforms and media: 4
- Welfare and public services: 3
- Elections and public information: 1
One mark is one record.
The records so far.
16 records
Delhi Police face recognition lists people who were in jail as present at a protest
Face or voice recognition. Failed or made an error.
New DelhiCourts, tribunals and policing
Deepfake eye-blink videos allegedly used to pass Aadhaar face checks and take loans in other people’s names
Synthetic media. Used deliberately to cause harm.
Ahmedabad, GujaratFinance and payments
Grok used to make sexualised images of women from their photos, prompting a government notice to X
Synthetic media. Used deliberately to cause harm.
Online, across IndiaOnline platforms and media
₹425 crore customs penalty set aside because the order rested on fabricated case citations
Language model or chatbot. Failed or made an error.
Surat, GujaratCourts, tribunals and policing
Mandatory face authentication blocks take-home rations for women and children at anganwadis
Face or voice recognition. Worked as built, and the way it was used caused harm.
Across IndiaWelfare and public services
AI-generated sexual images of a woman posted under a fake persona that drew over a million followers
Synthetic media. Used deliberately to cause harm.
AssamOnline platforms and media
Income tax tribunal recalls an order that cited four judgments which do not exist
Language model or chatbot. Failed or made an error.
Bengaluru, KarnatakaCourts, tribunals and policing
Deepfake videos of the RBI Governor used to promote investment schemes
Synthetic media. Used deliberately to cause harm.
Across IndiaFinance and payments
AI voice clones of two Bollywood actors used in fake election videos
Synthetic media. Used deliberately to cause harm.
Across IndiaElections and public information
Deepfake videos of the National Stock Exchange’s chief executive give stock tips
Synthetic media. Used deliberately to cause harm.
Across IndiaFinance and payments
Deepfake of Sachin Tendulkar used to advertise a gaming app
Synthetic media. Used deliberately to cause harm.
OnlineOnline platforms and media
Deepfake video of the actor Rashmika Mandanna spreads on social media
Synthetic media. Used deliberately to cause harm.
OnlineOnline platforms and media
Deepfake video call impersonating a former colleague used to take ₹40,000
Synthetic media. Used deliberately to cause harm.
Kozhikode, KeralaFinance and payments
Family ID database in Haryana records living pensioners as dead and stops their pensions
Automated decision system. Failed or made an error.
HaryanaWelfare and public services
Delhi Police treats face recognition matches above 80% similarity as positive identifications
Face or voice recognition. Worked as built, and the way it was used caused harm.
DelhiCourts, tribunals and policing
Welfare algorithm in Telangana wrongly tags poor families as ineligible for subsidised food
Automated decision system. Failed or made an error.
TelanganaWelfare and public services
Most of these cases are reported once and then lost.
A deepfake fraud gets a few paragraphs in a city paper. An algorithm that cuts off welfare payments is written up by an investigative outlet, and a chatbot’s mistake by a technology site. Each story stands on its own, which makes it hard to see which failures keep repeating, where they cluster, or whether anything changed afterwards.
International databases hold some Indian cases, but they aren’t organised around India: by sector, by state, or by what courts and regulators here did next. The register puts these cases in one place and one format. It starts with ones that are already well documented in public, and it grows as people report new ones. It isn’t an official record, and it can’t tell you how common these harms are, because it only holds what has been reported and what we could check.
Know of a case that isn’t here?
You can report something that happened to you, something you saw, or something you’ve read about. It takes about five minutes and you don’t have to give your name. Every report is checked against its sources before anything is published.
Use the data.
The whole register can be downloaded and reused with credit, under a Creative Commons Attribution licence. If you build on it or find a mistake in it, we’d like to hear from you.