NEW DELHI, Oct 16: India’s cybersecurity agency CERT-In has issued a critical warning about the misuse of artificial intelligence agents, highlighting how emerging AI tools could expose corporate information and compromise digital infrastructure if deployed without adequate safeguards.
In an advisory titled AI Agents as Attack Vectors, the Indian Computer Emergency Response Team warned that threat actors are exploiting open-source agentic AI tools to conduct cyberattacks against information and communication technology infrastructure. According to the agency, these tools can accelerate malicious operations, reducing the time needed to carry out certain attacks from days to hours.
The advisory highlights two categories of incidents involving unauthorised access to sensitive corporate information and live systems. The warning underscores the growing challenge faced by organisations as AI applications acquire the ability to perform tasks, interact with software platforms and access digital resources with limited human intervention.
Unlike conventional chatbots, which primarily respond to user instructions, AI agents can be designed to execute multi-step operations. Depending on their permissions and configuration, they may interact with files, databases, business applications and online services. These capabilities can improve efficiency, but they also create additional opportunities for misuse when security controls are inadequate.
CERT-In’s warning comes as companies increasingly experiment with autonomous systems to automate administrative work, software development, customer support, data analysis and other business functions. The same capabilities that make these systems attractive to businesses could also be exploited by malicious actors seeking to access confidential records or disrupt operations.
AI agents create new security challenges
Traditional cybersecurity strategies often focus on protecting networks, user accounts, endpoints and applications from unauthorised access. The introduction of AI agents adds another layer of complexity because these systems may be authorised to carry out actions on behalf of employees or organisations.
An agent connected to several internal services could potentially retrieve documents, process information and initiate actions across different platforms. If its permissions are excessive, its instructions are manipulated or its credentials are compromised, the resulting damage could extend beyond the application in which it operates.
The risks are not limited to sophisticated attacks. Poorly configured access controls, inadequate monitoring and excessive reliance on automated decisions can create openings that attackers may exploit.
For example, an AI assistant used to summarise company documents might have access to confidential files that are not necessary for its assigned task. If an attacker succeeds in manipulating its behaviour or exploiting a weakness in its environment, sensitive information could be exposed.
Similarly, an agent integrated with business software might be capable of performing actions that affect customer records, financial information or operational systems. Organisations therefore need to assess not only the security of the AI model itself but also the permissions and external services available to it.
CERT-In’s advisory draws attention to the need for businesses and AI developers to treat autonomous tools as part of their wider cybersecurity environment rather than as isolated productivity applications.
Faster attacks increase pressure on security teams
The use of AI in cybercrime has raised concerns about the speed at which attackers can identify weaknesses, automate repetitive tasks and coordinate operations. Activities that previously required considerable manual effort can increasingly be supported by automated tools.
CERT-In warned that the weaponisation of open-source agentic AI could compress cyberattack timelines substantially. This creates additional pressure on security teams, particularly those responsible for protecting large networks with limited personnel and resources.
When attacks move faster, organisations may have less time to identify unusual behaviour, investigate suspicious activity and contain a breach before it spreads. Security teams must consequently improve their ability to detect threats continuously rather than relying exclusively on periodic audits or manual reviews.
However, faster automation does not mean every AI-enabled attack will succeed or that all agentic systems are inherently dangerous. The level of risk depends on the tools being used, their permissions, the target environment and the effectiveness of existing safeguards.
The challenge for defenders is to understand how these systems behave under normal conditions and identify actions that fall outside established patterns. Unusual access to sensitive files, unexpected connections to external services and unauthorised changes to business systems may require immediate investigation.
Companies urged to strengthen access controls
The advisory places renewed emphasis on the responsibility of organisations deploying AI tools to establish clear security boundaries. Businesses should determine what information an agent can access, which actions it can perform and when human approval is required.
One important measure is the principle of least privilege. Under this approach, an AI agent receives only the permissions necessary to complete its assigned function. Access should not automatically extend to every document, database or application available to the employee or department using the tool.
Companies should also separate sensitive systems from general-purpose AI environments wherever practical. Restrictions on external connections, careful management of credentials and the use of controlled execution environments can help limit the consequences of a compromise.
Monitoring is equally important. Organisations need visibility into the actions taken by AI agents, including the information they retrieve, the services they contact and any changes they make. Logs should be protected against tampering and reviewed for suspicious activity.
Businesses should also establish procedures for immediately suspending an agent, revoking its credentials and isolating affected systems when a potential breach is detected.
Regular security testing can help identify weaknesses before attackers exploit them. These assessments should examine not only conventional software vulnerabilities but also risks arising from manipulated instructions, unsafe integrations, excessive permissions and unintended actions.
Developers must build security into AI systems
AI developers also face responsibilities in reducing the risks associated with autonomous applications. Security measures should be incorporated during design and testing instead of being added only after a product has been deployed.
Developers should restrict access to sensitive tools, validate requests before executing high-impact actions and prevent untrusted information from overriding authorised instructions. Systems handling confidential data should have clear rules governing what information can be retrieved, processed or transmitted.
High-risk operations, such as deleting important records, changing access permissions or transferring sensitive information, may require additional verification or explicit human approval.
Testing should include scenarios in which an agent encounters misleading instructions, malicious documents or compromised external services. Such exercises can help developers identify whether the system remains within its authorised boundaries under adverse conditions.
Organisations adopting third-party AI products should also examine the security practices of their providers. Understanding how data is stored, processed and retained is essential when systems handle commercially sensitive or personally identifiable information.
A growing concern for India’s digital economy
India’s expanding digital economy relies on interconnected networks across banking, telecommunications, healthcare, government services, manufacturing and information technology. A serious cybersecurity incident affecting one organisation can have consequences for customers, business partners and connected service providers.
As more companies adopt AI-powered automation, security policies will need to evolve alongside technological capabilities. Organisations that introduce autonomous tools without evaluating their permissions and operational risks could unintentionally increase their exposure to data breaches.
Smaller businesses may face particular difficulties because they often lack dedicated security teams and may rely on external providers to manage their IT infrastructure. Clear deployment guidelines, employee training and affordable monitoring tools can help reduce these gaps.
The latest warning also reinforces the importance of employee awareness. Staff should understand which AI tools are approved for workplace use and avoid submitting confidential information to applications that have not been assessed by their organisation.
Balancing innovation with protection
AI agents offer substantial opportunities to improve productivity by handling routine work, assisting technical teams and helping businesses process information more efficiently. Their long-term value, however, will depend partly on whether organisations can deploy them without exposing critical systems to unacceptable risks.
Companies should establish clear accountability for automated actions, maintain human oversight where necessary and regularly review the permissions granted to AI applications. Security teams must also adapt their detection and response procedures to account for automated activity.
CERT-In’s advisory serves as a reminder that cybersecurity strategies must keep pace with the changing capabilities of digital tools. As AI becomes more deeply integrated into corporate operations, protecting information will require a combination of technical controls, responsible development, employee awareness and continuous monitoring.
The central challenge is to ensure that automation strengthens business operations without creating new pathways for unauthorised access. For Indian organisations, treating AI security as an essential part of digital transformation will be increasingly important as adoption expands.