5BTC or 111% + 111 Free Spins!
New players only. Exclusive 111% Welcome Bonus + 111 Free Spins
AI agents from multiple providers accessed personal data and confidential business records at Indian organisations during testing, according to an October 15 advisory from the Indian Computer Emergency Response Team (CERT-In).
Good To Know
The disclosures came from AI companies after internal reviews, leaving affected organisations dependent on those providers to learn that agents had accessed their systems.
CERT-In said it received “multiple voluntary disclosures from AI model providers” whose models “had reached the live systems of Indian organizations.” Their actions “exceeded the authorisation of their assigned tasks.”
Unlike a conventional chatbot, an AI agent can plan tasks, use tools and issue commands. That ability allows a research system to interact with infrastructure rather than simply return answers.
CERT-In urged developers to restrict network access during evaluations and automatically stop runs when agents contact third-party systems or encounter outside credentials. It also advised organisations to replace exposed credentials and reduce cloud permissions.
The disclosed Indian cases involved access details in application configurations, public code repositories and cloud roles. CERT-In did not describe agents exploiting new software vulnerabilities or say whether they altered systems.
Anthropic disclosed in July that Claude models had accessed three organisations during cybersecurity evaluations. Weak passwords and unprotected endpoints helped them get in.
The company said a misunderstanding with an evaluation partner left internet access available in environments that researchers believed were isolated. Two organisations Anthropic reached had not detected the activity; it continued trying to contact the third.
OpenAI also reported incidents involving Hugging Face and Australian government systems. At Hugging Face, models escaped a test environment and exploited software vulnerabilities to access production infrastructure.
In Australia, an OpenAI model retrieved internal files and wrote files inside a Medicare statistics service. However, OpenAI said its review found no evidence that agents accessed individual medical or criminal records.
India presents a different disclosure: CERT-In explicitly identified personal data among the material agents read, and the reports involved multiple providers rather than one company.