India plans to log AI meltdowns in new database

Artificial intelligence (AI) is no longer confined to research labs or sci-fi thrillers—it’s powering telecom networks, running predictive systems in critical infrastructure, and helping decide everything from call routing to traffic flow. But what happens when it goes wrong?

India is preparing for that very question.

The government has released a draft framework to systematically record AI failures across the country’s telecommunications and digital infrastructure sectors. The move is part of a broader push to bring structure, transparency, and accountability to how AI is monitored and governed. Think of it as a national logbook for AI meltdowns—one that doesn’t just capture when things break, but also asks why they did and how bad the fallout was.

The proposed standard, known as TEC 57090:2025, has been developed by the Telecommunication Engineering Centre (TEC), the technical arm of the Department of Telecommunications.

It lays out a schema—that is, a structured template—for how AI incidents should be documented. Alongside this, it introduces a taxonomy: a formal way of categorizing the kind of damage AI might cause, from a system glitch to a full-blown privacy breach.

Why now?

As AI technologies continue to be embedded in everything from mobile networks to smart meters, the potential for things to go wrong—whether through malfunction, misconfiguration, or misuse—is growing. Already, there have been scattered reports globally of chatbots going rogue, facial recognition tools misidentifying people, and predictive systems triggering unintended consequences. But India wants to stop playing catch-up.

Under the draft, every AI incident reported will be recorded with standardized fields: a title, a short summary, the date and location, the AI system involved, who built it, who deployed it, what its intended purpose was, and what exactly went wrong.

Was it a network blackout caused by a faulty prediction model? Or a breach where an AI mistakenly granted someone access to restricted data? Each scenario will have its place in the database.

The idea is to not just log incidents but make them searchable and comparable. That’s where the taxonomy comes in.

Say an AI failure causes a power outage, or a telecom glitch leads to a service collapse in a rural area.

Was it because the AI system was badly configured? Was it a cybersecurity loophole that got exploited? Or did the developers overlook a design flaw? The new taxonomy breaks these scenarios down by type—network disruption, security breach, AI mismanagement—and by impact, such as physical harm, financial loss, reputational damage, or violation of privacy.

Even the severity of the incident will be graded, from minor hiccups like slow data speeds to nationwide outages or legal consequences. In short, every AI incident will come with its own digital case file.

But who’s expected to fill these in?

The framework is designed with multiple users in mind. Developers and telecom operators, of course, are key—they’ll be expected to record issues during testing and after deployment. Regulators and policy-makers will use the data to make more informed rules. Academic researchers will mine the records for trends and case studies. And then there are the incident reporters: employees, users, or even whistleblowers who spot something wrong and want to report it.

To keep things credible, each submission must include contact details, the source of the information (such as a news article or internal memo), and an optional section for extra context. This ensures that reports can be verified without discouraging transparency.

While the current focus is on telecom and critical digital infrastructure, the implications could stretch far wider. The same model could, in the future, apply to sectors such as finance, transport, or healthcare—anywhere AI is used in ways that affect large populations or sensitive services.

Meanwhile, this isn’t India’s first step into AI governance.

The country has already launched initiatives to boost AI innovation, but this is among the first serious moves to build a public accountability layer around it. And it comes as other countries begin similar efforts.

The European Union’s AI Act, for example, classifies high-risk AI applications and requires post-market monitoring. The US, through its NIST framework, is also nudging companies toward transparent risk management.

NIST, or the National Institute of Standards and Technology, is a US government agency under the Department of Commerce that has developed a risk management framework that helps organizations identify, assess, manage, and monitor risks associated with AI systems.

What makes India’s approach different is its attempt to standardize not just how incidents are logged, but how they are understood across technical, regulatory, and ethical dimensions.

The goal, TEC noted, is not merely to flag problems after they happen—but to create a system that helps prevent them in the first place.

The draft is open for public consultation until 27 July.

TEC has invited comments from stakeholders, including industry bodies, telecom operators, civil society groups, and individual experts. Suggestions must follow a prescribed format, which includes the specific clause being commented on, the proposed revision, and the justification behind it.

If adopted, the standard could form the basis for national AI incident reporting in India—and perhaps even influence international norms. For now, the goal is simple: log the breakdowns before they break public trust.