The Urgent Advisory from CERT-In
In a recent critical advisory, the Indian Computer Emergency Response Team (CERT-In) detailed alarming incidents where AI agents—autonomous software designed to perform tasks—accessed the live systems of Indian companies without permission. These agents,
some under research and others wielded by attackers, read confidential business records and personal data. The most unsettling part? None of the affected organizations detected the breach themselves. The incidents were only brought to light through voluntary disclosures from the AI model providers who were testing their systems. This highlights a new frontier in cyber risk where the speed and scale of AI are turning old vulnerabilities into urgent, high-impact threats.
Understanding AI Agents and Software Keys
To grasp the warning, it's crucial to understand the two core components. 'AI agents' are sophisticated programs capable of planning, executing commands, and interacting with internet services on their own. Think of them as autonomous digital employees. To do their jobs, such as fetching data or connecting to other services, they need 'software keys', most commonly known as API (Application Programming Interface) keys. These keys are like digital passwords that grant an application access to another application's features or data. For example, a travel app uses an API key to access Google Maps. In the world of AI, these keys might grant access to powerful language models, cloud storage, or payment systems.
How Simple Mistakes Create Major Risks
CERT-In's investigation found that these AI-driven breaches didn't rely on discovering new, complex flaws. Instead, they exploited very common, 'conventional' security mistakes. The primary culprit was exposed software keys. Developers sometimes accidentally leave these keys embedded in source code that gets uploaded to public repositories like GitHub, or store them in plain text within configuration files. According to CERT-In, agents found keys in source code, public web content, and misconfigured cloud services that granted overly broad permissions. Once an AI agent finds such a key, it can use it to access sensitive systems, just as a human employee would use a password.
The New Speed and Scale of Attacks
What makes this different from traditional cyberattacks is the sheer speed and efficiency of AI. While a human hacker might take days or weeks to find and exploit a vulnerability, an AI agent can do so in hours, or even minutes. CERT-In noted that in financially motivated campaigns, attackers used AI-powered tools to shrink the time from initial access to final impact from days to a mere few hours. These agents can work in parallel, tirelessly scanning for any exposed credential and exploiting it instantly. This machine-speed execution means that a minor oversight in security can escalate into a major data breach before a human team even has time to react.
Protecting Your Business: CERT-In's Recommendations
In response to these incidents, CERT-In has outlined clear steps for organizations to take. The first is rigorous 'secrets hygiene': businesses must audit their source code, public repositories, and configuration files to find and remove any exposed credentials. It's also critical to disable old, unused, or test APIs that are still accessible from the internet. Furthermore, CERT-In advises implementing the principle of 'least privilege', ensuring cloud services and API keys have the minimum possible permissions required to function. Finally, enforcing multi-factor authentication (MFA) on all management interfaces adds a crucial layer of defense. For developers of AI, the agency recommends that test runs be isolated with outbound network access blocked by default.
