IT Home On October 16, according to a report by Reuters this afternoon (16th), the Indian Computer Emergency Response Team (CERT-In) revealed that some AI models unauthorizedly invaded the business systems of several Indian institutions during testing, reading confidential information, and the affected institutions were not aware of the entire process.
This is the first time that India has reported such a situation after similar AI overreach incidents occurred in other parts of the world.
The report states that several AI model suppliers have voluntarily reported that their AI agents exceeded the authorized scope during testing, and accessed data such as business records of Indian companies.
The institution has not revealed the names of the involved AI company and the affected institutions.
This incident also brought India into global discussions regarding AI model security testing, with the focus being on how AI companies should test models with strong network intrusion capabilities.
According to IT Home, in July this year, OpenAI revealed that its models went out of control during testing and invaded a startup. Anthropic also reported that there were at least four incidents this year where its AI models invaded external systems. OpenAI has also apologized for its失控 AI agent invading the Australian government website.
The Indian Computer Emergency Response Team stated that none of the affected Indian institutions discovered the intrusion on their own, and the incident was reported by an AI company.
The agency also found in the investigation of another type of incident that, in order to obtain economic benefits, the attackers deliberately used AI tools that were available for free to probe India’s computer systems, in search of opportunities to carry out intrusions. Such tools are also widely deployed in the infrastructure of multiple countries.
The Indian Computer Emergency Response Team said that it used to take several days for a network attack from the initial intrusion to cause damage, but now it has been reduced to a few hours. If the affected organization fails to completely eliminate the security threats, it may be invaded again within a few days, and the attackers could further penetrate the system.