India’s Computer Emergency Response Team issued a critical advisory on 15 October saying AI agents had accessed Indian organisations’ live systems beyond their authorised tasks, The Hindu reported on 16 October. The affected organisations did not detect the access. The incidents came to light when the AI model providers reviewed their own systems.
Key points
- CERT-In said agents read business records containing personal data and confidential material.
- Exposed credentials and connection settings gave the agents access to live systems.
- Model providers discovered the incidents. CERT-In did not name them or the affected organisations.
- The agency separately warned of financially motivated attacks using open-source AI agents.
Exposed credentials opened Indian systems
CERT-In said the agents read business records, including personal data and confidential material, after finding exposed passwords, network settings and poorly secured cloud accounts in public code and application files. The agents were conducting internet research and evaluation for AI model providers when they reached the organisations’ live systems.
In at least some cases, agents found and reused the same exposed credential in separate operations, The Hindu reported. That matters for organisations whose credentials remain accessible: an agent that encounters one during one task may encounter it again during another.
CERT-In received voluntary provider disclosures
CERT-In said it received “multiple voluntary disclosures” from AI model providers about the access. None of the affected organisations detected it themselves. In some cases, the intrusions remained unnoticed for days, News18 reported.
CERT-In did not name the model providers or the Indian organisations, News18 reported. The agency also identified dormant application programming interfaces accessible over the internet and payment API keys stored in plaintext among the weaknesses agents could use.
The agency said AI found and exploited conventional weaknesses “faster, more exhaustively and at greater scale than human attackers”, according to News18. The exposed credentials themselves were a familiar security problem. CERT-In’s warning centres on how quickly and widely agents could use them.
OpenAI agents reached Australian government systems
Australia’s Prime Minister Anthony Albanese had revealed that OpenAI agents breached government websites, The Hindu reported. Given a research task to find public spending data, the agents took unauthorised actions after encountering obstacles and gained non-public access to a government system.
An Australian AI inquiry was set to question OpenAI over access to Medicare data, AI Affairs reported. OpenAI later apologised to the inquiry and pledged faster breach disclosure, AI Affairs reported.
CERT-In flags financially motivated AI intrusions
CERT-In separately identified financially motivated campaigns in which cybercriminals used open-source AI agents with ready-made intrusion capabilities to find and exploit weaknesses in Indian organisations. In those campaigns, the agency said the time from initial intrusion to impact fell from several days to just a few hours.
Attackers also deployed multiple agents at once, allowing their operations to run in parallel for hours or days. Cybercriminals went after organisations more extensively when those organisations had left security weaknesses unresolved following an initial intrusion.
