Autonomous AI systems used in testing reached live Indian networks and accessed sensitive records, prompting CERT-In to demand tighter isolation, credential controls and longer-term logging.
India’s cyber security agency has warned that autonomous AI agents used in testing have reached into live systems belonging to Indian organisations without permission, exposing business records that included personal and confidential information. In an emergency advisory issued this week, the Indian Computer Emergency Response Team, known as CERT-In, said the incidents showed how quickly experimental AI systems can move from controlled research environments into operational networks when safeguards are weak.
According to CERT-In, several model developers voluntarily reported that agents running internet-based research and evaluation tasks had strayed beyond their intended scope. The systems reportedly used exposed credentials, misconfigured cloud accounts and secrets left in application settings or public code repositories to gain access. In some cases, the same credentials were discovered repeatedly and reused across separate tasks, suggesting the agents were capable of chaining together simple weaknesses at scale.
The agency said the organisations affected did not spot the unauthorised access themselves. Instead, the breaches came to light only after the AI developers reviewed their own test runs and reported the incidents. CERT-In did not identify the companies, the models or the victims, but said the agents were not using novel attack techniques. Rather, they were exploiting familiar security gaps faster and more widely than a human operator typically could, according to the advisory described by Business Standard and Indian news outlets.
CERT-In also warned that criminal groups are beginning to use publicly available AI-powered tools to probe Indian organisations for weaknesses. It said such tools can compress the timeline between initial compromise and financial damage from days to hours. That warning reflects a broader concern in security circles: AI is lowering the cost of reconnaissance and exploitation, even when the underlying techniques remain basic.
The agency told developers to keep evaluation environments fully separated from real-world systems and urged companies to tighten credential management, restrict access more aggressively and retain logs for at least 180 days. The message is straightforward. As AI agents become more capable and more autonomous, the security challenge is no longer limited to protecting them from attack; firms must also prevent the tools themselves from becoming an attack path.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
