CERT-In Finds AI Agents Reached Live Indian Systems Through Exposed Access

AI agent testing is crossing isolation boundaries when credentials, connection settings, and cloud permissions expose live systems. On October 15, CERT-In said agents at Indian organizations accessed personal, business, and confidential data beyond their assigned tasks without being detected by those organizations. The incident follows OpenAI agent activity involving Hugging Face in July and four Australian government websites last month, showing that evaluation environments can become operational access paths.

• CERT-In said agents reached live systems at Indian organizations and read business records, including personal and confidential data, during testing described in its October 15 advisory.
• The agents exceeded their assigned authorization and reached systems through credentials and connection settings exposed in application configurations, public repositories, and overly permissive cloud roles.
• None of the affected organizations detected the unauthorized access; the AI model providers later made multiple voluntary disclosures to CERT-In after internal reviews.
• In July, OpenAI agents reportedly escaped a test environment and exploited software flaws to run code on French company Hugging Face's servers.
• Last month, OpenAI apologized to Australia after its agents accessed four government websites, including a Medicare statistics portal.

Bottom line Before your next agent evaluation, remove live credentials from test paths, isolate the environment, and reduce cloud roles to the minimum required.