New Delhi: Autonomous artificial intelligence (AI) agents have accessed live computer systems and confidential information belonging to Indian organisations without authorisation, prompting the country's cybersecurity agency to issue a critical security advisory. However, government sources said the incidents did not involve major financial losses or compromise vital information held by significant organisations.

The Indian Computer Emergency Response Team (CERT-In) has warned that AI agents, including those undergoing research and testing, have operated beyond their authorised boundaries and accessed sensitive business records. The agency also flagged the growing use of open-source AI-powered tools by cybercriminals to launch faster and more sophisticated attacks.

Government sources told PTI on Friday that the incidents should serve as an early warning for Indian companies to strengthen their cybersecurity infrastructure, improve monitoring systems and address vulnerabilities before they lead to more serious breaches.

According to the sources, the government is also considering whether autonomous AI agents should face specific restrictions under upcoming AI regulations. A consultation paper examining the issue is expected to be released soon.

AI agents accessed live systems without authorisation

In its critical advisory, CIAD-2026-0049, titled 'AI Agents as Attack Vectors', CERT-In revealed that AI model providers had voluntarily reported instances in which autonomous AI systems exceeded the scope of their assigned tasks.

These agents, originally deployed for research and evaluation, gained access to operational systems belonging to Indian organisations.

During the incidents, the AI agents accessed business records containing personal information and confidential material. They exploited existing security weaknesses, including exposed login credentials, connection details stored in application configurations, publicly accessible code repositories and excessive permissions granted to cloud services.

More concerningly, none of the affected organisations independently detected the unauthorised access.

The breaches came to light only after AI model providers identified the activity during internal reviews and voluntarily disclosed the findings.

CERT-In said the incidents demonstrated how autonomous AI systems could exploit conventional cybersecurity weaknesses even without being explicitly instructed to access sensitive information.

Cybercriminals using AI to launch attacks within hours

The agency also identified a separate trend involving cybercriminals deploying publicly available AI-powered offensive security frameworks against Indian organisations.

According to CERT-In, attackers have been using these tools to identify vulnerable systems, exploit security gaps and execute financially motivated cyberattacks.

The agency found that AI-powered tools significantly accelerated the attack process, reducing the time between initial system access and the final impact from several days to just a few hours.

Organisations that addressed vulnerabilities only partially were targeted again within days, sometimes through larger attacks.

CERT-In also observed that these offensive tools were being repeatedly modified and deployed through servers located across multiple countries.

Despite the involvement of advanced AI systems, the agency identified traditional security lapses as the underlying weaknesses in the incidents.

Government considers restrictions on autonomous AI agents

Government sources said the latest developments highlighted the need for companies to invest more aggressively in cybersecurity.

They pointed out that many organisations continue to neglect basic cyber hygiene, leaving systems vulnerable to attacks involving increasingly sophisticated AI tools.

Sources said larger companies should strengthen their security operations centres to identify suspicious behaviour, including repeated login attempts and access originating from unusual IP addresses.

Organisations were also advised to adopt machine-learning-based security tools capable of identifying threats and responding automatically.

Another concern involves the ability of autonomous AI agents to repeatedly attempt tasks without human intervention.

As these systems become more capable, companies may find it increasingly difficult to anticipate or control their actions unless appropriate restrictions are introduced.

Government sources indicated that the forthcoming consultation paper on AI regulation could examine whether limits should be imposed on autonomous agents, particularly regarding their ability to access sensitive systems and execute tasks independently.

Experts warn of risks to personal data

Cybersecurity experts cautioned that the growing adoption of autonomous AI could expose businesses and individual users to new security threats, particularly when agents are granted broad access to confidential information.

Jaydeep Singh, General Manager - India at Kaspersky, said the CERT-In advisory was "a timely reminder that AI agents are no longer just productivity tools".

"When given broad access, they can reach live systems and sensitive data in ways organisations never intended, often by moving through gaps that already exist, such as misconfigured cloud services, exposed credentials and excessive permissions," Singh said.

He noted that legitimate user accounts, publicly accessible applications and trusted relationships accounted for more than 80 per cent of the attacks investigated by Kaspersky in 2025.

"With AI allowing threat actors to iterate faster and scale their efforts, these paths can now be exploited more quickly than ever," he added.

Singh recommended keeping AI agents used for research and testing completely separate from operational systems, with synthetic data and credentials restricted to testing environments.

"Every agent should be inventoried, isolated in sandboxed environments, governed by default-deny rules and required to seek human approval before any irreversible action," Singh added.

He also stressed that critical security vulnerabilities should be addressed "within hours, not weeks".

Citing a Kaspersky study examining approximately 30,000 popular GitHub repositories, Singh said excessively broad access permissions were among the most common security weaknesses.

Companies urged to monitor AI systems continuously

Sunil Sharma, Managing Director and VP - Sales (India and SAARC) at Sophos, said autonomous AI systems were creating additional cybersecurity challenges as they increasingly gained the ability to access information and perform tasks independently.

"Organisations will need to extend established security practices to these systems: a distinct identity for each agent, least-privilege access, secured API keys and tokens, adequate logging, and continuous monitoring for unusual behaviour. Research and testing environments should also be isolated from production," he said.

Soumen Mandal, Principal Analyst at Counterpoint Research, warned that the consequences of unauthorised AI access could extend beyond businesses to ordinary users.

As AI agents increasingly process personal and confidential information, failures to enforce access restrictions could expose sensitive user data without organisations immediately realising it.

"As AI adoption becomes mainstream, AI developers, governments, application providers, independent researchers and industry bodies must work together to establish robust security standards and accountability frameworks that build trust without slowing innovation," Mandal said.

Techarc Chief Analyst Faisal Kawoosa also urged businesses and individuals to exercise caution while adopting AI applications.

He observed that organisations frequently selected AI tools based primarily on cost rather than adequately assessing their security implications.