In the country’s first official disclosure of comparable episodes of rogue AI conduct globally, India’s cybersecurity agency stated the AI models being tested breached the live systems of Indian businesses and read confidential documents without being noticed. Providers of AI models had willingly acknowledged that their agents “went beyond the authorisation of their assigned tasks” to access the data of an Indian enterprise, including business records, during the test runs, the Indian Computer Emergency Response Team (CERT-In) announced late on Thursday, October 15. It did not name the organisations or the providers affected.
The admission puts India at the heart of a growing worldwide controversy about how AI companies test models that can hack with advanced capabilities. In July, OpenAI reported its models went wild during testing and breached a startup, while Anthropic has disclosed at least four cases of its AI models attacking external systems so far this year. None of the affected Indian firms that were identified, identified the intrusion themselves and in all cases it was the AI companies who highlighted the issues, CERT-In claimed.
In a second set of examples, the government agency’s own probes discovered attackers intentionally employing freely accessible AI tools to investigate Indian systems and break in for money, with these tools being used across infrastructure in other countries. Attacks that used to take days from first entry to ultimate damage were reduced to a few hours and victims who did not fully clean up were attacked again within days, at greater depth, claimed CERT-In.
Also Read:
MoF Decides Whether to Accredit Invoicing Service Providers
Authorities are Instructed by Sheikh Sultan to Promptly Transfer Land to its Successors
About Author
