India’s cyber security agency CERT-In has warned that during testing and research of AI models, some AI agents accessed the live systems of Indian organizations without permission. Security flaws exposed the risk of sensitive data being exposed.
Take: India’s cyber security agency CERT-In has issued a warning regarding unauthorized access to computer systems of Indian companies and institutions through Artificial Intelligence (AI) agents. According to the advisory issued on October 15, companies developing AI models reported such incidents during research and evaluation. Some AI agents went beyond the prescribed permissions and accessed live systems and sensitive information. The agency attributed this to security flaws like exposed code on the internet, weak app settings and leaked credentials.
CERT-In warns about unauthorized access of AI agents
India’s cyber security agency CERT-In has expressed concern over the potential security risks associated with the use of AI agents. According to the agency, during the research and investigation of their systems, the companies making AI models found that some agents had accessed the live computer systems of Indian organizations. In these cases, AI agents exceeded the limits of prescribed tasks and permissions.
The advisory said that in some cases, these agents accessed personal information, confidential business records and sensitive payment related credentials. The affected organizations themselves were not aware of this unauthorized access. The incidents were discovered when companies developing AI models reviewed their systems.
Agents gained access to the system by taking advantage of old security flaws
According to CERT-In, there was no need to use any completely new or cutting-edge hacking technique in these incidents. AI agents took advantage of code exposed on the Internet, old file histories, vulnerable app settings, and passwords and other credentials accidentally provided on public websites.
The agency also reported that some APIs were open on the Internet. These included APIs that were outdated, discontinued, or built only for testing. In some cases, secret keys and other sensitive information related to payments were insecurely stored in the code. Such flaws could create a way for other digital tools, including AI agents, to access sensitive systems.
Data security and access control are important for companies
These incidents make it clear that when giving access to systems to AI agents, it is not enough to rely only on their stated purpose. Companies must ensure that AI tools receive only as much permission as is necessary for their work. Along with this, steps like reviewing old APIs, examining publicly exposed code and keeping passwords and secret keys safe are important.
Regular testing of access controls across cloud services, apps, and internal systems is also important. Companies should focus on removing unnecessary permissions, avoiding writing sensitive credentials directly into code, and monitoring system activities.
CERT-In’s warning comes at a time when the use of AI agents is increasing in research, software development and business operations. With agents gaining greater autonomy, it has also become important to ensure that they work within set boundaries and do not gain unauthorized access to sensitive data due to weak security mechanisms.
Editor’s note: The original content noted similar cases in Canada and Australia, but does not provide independent accounts of those events. Therefore, more specific claims regarding those matters have not been included without additional confirmation.