Pune uses ChatGPT on CCTV footage to catch customers who don't pay: the problem is the "defaulter" label

By Urban Acres · October 25, 2026.

Some restaurants in Pune (India) are using AI for more than security: checking whether a customer has paid. According to Punekar News, which spoke with owners, an establishment in Parvati Paytha uploads footage from its CCTV (closed-circuit television) to ChatGPT. It wants to identify a person suspected of not paying and generate a sharper image of that "defaulter". Another owner said he has barely used it because his customers are honest, but that it could help determine who has paid during peak hours.

The evidence is limited and should be kept in proportion. It rests on the accounts of a few owners. The article, published by the think tank Urban Acres based on that report, does not say how many businesses do this. Nor does it provide data on the method's accuracy or say that the police or municipal authorities endorse it. The comments themselves suggest the practice is being explored, not that it is a verified payment system.

It is also unknown what happens to the images. The material does not clarify how long they are kept, who has access to them or whether customers are notified. Nor does it clarify whether they are deleted once the dispute is resolved or whether someone flagged as a defaulter can contest it. India's personal data protection law (Digital Personal Data Protection Act, 2023) and its 2025 rules are mentioned, but without analyzing how they apply to this case. It therefore cannot be said whether the practice breaches or complies with the law.

Our reading: what matters is not the technical novelty but the lowering of barriers. A restaurant does not need to build a surveillance network or a technology department. The cameras it already has and a general-purpose AI service are enough. That democratizes the tool, but also takes it to places where no one has thought about governance. Broadly, commercial surveillance tends to grow this way: through small experiments that become routine before any rules exist.

The central risk is that an uncertain observation turns into a reputational label. A packed restaurant produces obstructed views, incomplete footage and confusion. Someone may leave while another member of their group pays, or pay digitally after getting up. An AI-enhanced image is not proof of non-payment, and generating a "clearer" face can create a false sense of certainty. If that image circulated among staff or between establishments, a dispute over a bill could become a permanent record about a person. The material does not say this has happened in Pune. But it is the scenario that justifies keeping an eye on the practice now.

There is also a shift in purpose. A camera installed to deter theft ends up investigating an unpaid bill. And the difference between being recorded and being assessed by an algorithm is precisely what the customer does not perceive: the camera is visible, the AI analysis is not.

This is a small case of a short-term pattern worth taking seriously: cheap AI expands the capacity to surveil long before the capacity to hold anyone accountable. In the long term, these same tools can make services more efficient and fair. But only if they come with the basics: clear notice to customers, a defined purpose, retention periods, access logs and a real way to challenge an error.

What to watch: whether more establishments adopt the method, whether they explain how they handle the images, and whether regulators or AI providers publish guidance for the commercial use of these tools.