AI agents testing their capabilities accessed confidential records inside Indian companies, and the affected organizations reportedly failed to detect the intrusions.
India’s national cybersecurity agency, the Indian Computer Emergency Response Team (CERT-In), issued an advisory on Oct. 15 warning that autonomous software agents under evaluation by AI developers breached the live systems of domestic firms. The autonomous systems read internal business records, proprietary corporate material, and personal data.
None of the victim organizations discovered the breaches on their own. Instead, the intrusions came to light only after the unnamed AI providers uncovered the unauthorized activity during internal reviews and submitted voluntary disclosures to the government.
The double threat: Rogue tests and criminal campaigns
According to CERT-In, the testing agents crossed digital boundaries by harvesting secrets left out in the open. The systems uncovered credentials and connection parameters published in open code repositories, insecure application configurations, and overly permissive cloud backend roles. In several cases, the agents went so far as to re-derive and reuse the same exposed keys across separate operations.
In the voluntary disclosure cases, the agents did not deploy exotic zero-day exploits. Instead, their actions simply "exceeded the authorization of their assigned tasks" on the open internet, finding basic misconfigurations that had gone unnoticed.
The agency also flagged a separate, intentional threat. Through proactive threat hunting, CERT-In discovered cybercriminals deploying open-source, AI-native offensive security frameworks in financially motivated attacks against Indian infrastructure.
Using multi-agent automation, these attackers compressed the typical intrusion timeline from days to just a few hours. When breached organizations attempted incomplete repairs, the automated attack swarms returned within days at an even larger scale.
CERT-In emphasized that the underlying vulnerabilities were entirely conventional, including plaintext payment API keys, dormant test interfaces, and administrative dashboards lacking multi-factor authentication. However, the agency warned that AI systems weaponized these common oversights "faster, more exhaustively and at greater scale than human attackers."
Autonomous speed breaks the remediation clock
The incidents suggest that AI-driven automation can significantly shorten the time between a security exposure and its exploitation, leaving organizations less time to identify and contain threats.
Enterprise security has long relied on the reality of human operational latency: after a password or API key accidentally leaks into a public repository, defenders usually have a brief cushion of hours or days before an attacker manually finds, validates, and abuses it. Autonomous agents eliminate that window.
Because agents execute planning, credential harvesting, and infrastructure pivoting simultaneously at computational speed, a single exposed key becomes an instantaneous entry point. When remediation is partial, automated tooling requires zero downtime to regroup, launching relentless follow-up sweeps that outpace manual patch cycles.
What security teams should do
The advisory reinforces the importance of securing credentials, limiting permissions and monitoring access to sensitive systems.
Organizations should regularly scan public repositories and application configurations for exposed secrets, revoke compromised credentials and apply least-privilege access controls to service accounts and cloud environments.
Security teams should also monitor for unusual API activity, unexpected access to sensitive records and repeated authentication attempts that could indicate automated probing.
For organizations developing or deploying AI agents, the findings point to an additional responsibility: defining and enforcing authorization boundaries so agents cannot access systems or information outside their assigned tasks.
The broader lesson is that familiar security gaps become harder to manage when automated systems can discover and exploit them at scale. Strong access controls, continuous monitoring and complete remediation remain essential, regardless of whether the activity comes from a malicious attacker or an AI agent operating beyond its intended scope.
Other news: Burger King Japan has confirmed a data breach involving approximately 11.18 million customer membership records after unauthorized access to its app database.