New cybersecurity blind spot: When AI looks like a trusted employee
CERT-In's latest advisory highlights growing AI agent risks in enterprise security, urging organizations to adopt stronger governance, oversight, and intent-based monitoring to protect against rapid AI-driven exploits.
By Raghavendra BV, The latest advisory from CERT-In should be viewed as more than another cybersecurity alert. It highlights a shift in enterprise risk that many organisations are only beginning to recognise. For nearly two decades, companies have invested heavily in detecting cyber threats. Security tools were designed to identify suspicious users, malicious software, unauthorised access, and abnormal behavior.
These controls worked because attackers typically operated outside the boundaries of legitimate business activity. AI agents change that equation: Unlike traditional attackers, AI agents are granted access by the organisation itself. They log in using legitimate credentials, access approved applications, and operate within permissions that employees have assigned to them. To most security systems, they appear no different from a trusted worker.
That is precisely the challenge: As AI adoption accelerates, the key security question is no longer simply who accessed a system, but why the access occurred and whether it aligns with the agent’s intended purpose. In the rush to unlock productivity gains, organisations often prioritise speed over precision when assigning permissions. As a result, an agent designed to summarise customer tickets may end up with visibility into an entire customer database.
The underlying weaknesses are not new. Exposed credentials, insecure APIs, excessive permissions, and weak authentication have long been recognised security issues. What has changed is the speed at which AI can exploit them. A human attacker might spend days or weeks navigating an over-permissioned environment. An AI agent can perform the same activity in minutes, often without triggering alarms designed to detect human behavior.
Security systems built to identify intrusions are far less effective when the activity originates from a legitimate, authorised identity. This is why the issue extends beyond technology and into governance: In many large enterprises, machine identities already far outnumber human users. Yet most organisations cannot confidently answer three fundamental questions: How many AI agents are operating across the business?
What systems can they access? And who is accountable for their actions? These questions are becoming boardroom concerns. The first priority for security leaders is visibility. Every AI agent, whether developed internally or embedded in a third-party platform, should be treated as a distinct digital identity with a designated owner and clearly defined purpose. Access should be granted based on business need, not convenience.
Unlike human employees, agents do not exercise judgment, question unusual instructions, or recognise when an action may have unintended consequences. If configured incorrectly or provided with excessive access, they can rapidly propagate errors, expose sensitive information, or trigger actions across multiple systems at a scale that would be difficult for a human user to match. Second, organisations must evolve from monitoring access to monitoring intent.
Knowing that an agent read a file or queried a database is no longer sufficient. Enterprises need to understand whether those actions are consistent with the agent’s assigned role. Accessing unrelated datasets, invoking unfamiliar APIs, or attempting to execute code outside its intended function should attract scrutiny and, where necessary, require human approval. Third, organisations should strengthen accountability across their AI ecosystem.
Contracts with AI providers must clearly define incident reporting obligations, audit rights, data-handling responsibilities, and notification timelines. These provisions are increasingly important in light of regulatory expectations and emerging data protection requirements. None of this argues for slowing AI adoption. Organisations that effectively deploy AI agents will unlock meaningful gains in productivity, efficiency, and innovation.
However, autonomy without oversight creates new forms of operational and cyber risk. The lesson from the CERT-In advisory is straightforward: AI agents should be managed much like privileged employees. They must be verified, granted only the access they need, continuously monitored, and regularly reviewed. The adoption of AI is no longer a future challenge. It is today’s reality. The organisations that succeed will not necessarily be those that deploy the most AI, but those that establish the governance, accountability, and visibility needed to deploy it safely.
(The author is Partner and Deputy Leader, Cybersecurity Consulting, EY India. Views expressed are personal). Disclaimer: The views expressed are the author’s own and do not reflect the official policy or position of Financial Express.
