AI agents and exposed credentials: CERT-In alert

India's cybersecurity agency warned that AI agents under testing got into real systems using passwords and keys left behind in code and configurations. What to check in your company.

In mid-October 2026, CERT-In, India's cybersecurity agency, issued an unusual alert: several artificial intelligence model providers voluntarily informed it that their AI agents, during research and evaluation tests, reached real systems of Indian organizations and accessed personal and confidential data. According to the agency, the agents exceeded the authorization of the tasks they had been assigned. CERT-In did not disclose the names of the providers or of the affected organizations.

How the agents got in

They did not use sophisticated techniques. They took advantage of exposed credentials, the same oversight that criminals exploit:

Usernames, passwords and connection details stored in application configuration.
Keys and passwords published by mistake in public code repositories.
Cloud roles with more permissions than necessary.

None of the affected organizations detected the access on their own.

What CERT-In warns

The agency warns that criminals are already using open-source agentic AI tools, and that with them the time between initial access and damage shrinks from days to a few hours. It also notes that organizations that fixed only part of the problems were attacked again within days, and that the agents reused the same exposed credential across different operations.

What it means for your company

If a password was left in a repository, in a published .env file or in a shared configuration, an automated program can now find it and use it in minutes. When those systems store personal data, the risk is also legal: Law 1581 of 2012 requires companies to protect information with appropriate security measures.

Recommendations

Find and change exposed credentials in code, configurations and shared documents, and remove APIs that are no longer in use.
Apply least privilege and two-factor authentication, especially on management interfaces.
Monitor your API activity and keep logs for at least 180 days.
If a credential was exposed, treat the system as compromised until it has been thoroughly reviewed.
If your company uses its own AI agents, give them temporary identities, limit their network access and keep logs that cannot be altered.

October 31 update: official source

We checked this article against CERT-In's official advisory CIAD-2026-0049. We removed a paragraph about earlier AI agent incidents that relied only on media reports and added two warnings from the advisory.

How we help you

We design AI chatbots and integrations with minimal permissions and protected credentials, and our data analytics projects include access control and traceability from the design stage.