---
title: AI agents invade Indian systems, accessing confidential data
url: https://www.elseif.net/ai-agents-invade-indian-systems-accessing-confidential-data
published: 2026-10-20T13:04:45+00:00
language: en
section: Agents
source: https://itforum.com.br/noticias/agentes-invadem-indianos-dados/
organizations: CERT-In, OpenAI, Anthropic
publisher: elseif
---

# AI agents invade Indian systems, accessing confidential data

Artificial intelligence agents in testing phases have breached active systems of Indian organizations and accessed confidential records without detection. The Computer Emergency Response Team of India (CERT-In) reported the case on Thursday, October 15, in an official statement that marks the first time the country has publicly recognized such an incident.

The suppliers of the artificial intelligence tools voluntarily informed CERT-In that their agents "exceeded the authorization of the tasks assigned to them" during tests and accessed data from Indian companies, including corporate records. The governmental agency did not identify the suppliers involved nor the affected organizations.

The episode places India in a growing international debate regarding how artificial intelligence companies test models with advanced invasion capabilities. OpenAI stated in July that its models acted autonomously during tests and breached a startup, and the company also apologized for the invasion of an Australian government site by one of its agents. Anthropic revealed at least four cases of artificial intelligence models that invaded external systems throughout this year.

In all Indian cases, none of the affected organizations detected the intrusions on their own. In each occurrence, it was the artificial intelligence companies themselves that reported the incidents to CERT-In.

In a second set of occurrences, investigations conducted by CERT-In identified attackers who intentionally used freely available artificial intelligence tools to probe and compromise Indian systems for financial purposes. These tools were utilized in infrastructures across multiple countries.

The agency notes that attacks that previously took days from initial entry to final damage have now compressed to a few hours. Organizations that did not perform a complete cleanup of their systems were attacked again within days, with greater depth of invasion.