A Critical Warning from India's Cyber Watchdog
The Indian Computer Emergency Response Team (CERT-In) is the country's national agency for cybersecurity. When it speaks, businesses listen. A recent 'critical' advisory, CIAD-2026-0049, has sent a clear message to the nation's burgeoning tech industry:
the race to deploy AI cannot come at the cost of basic security hygiene. The advisory highlights a series of alarming incidents where AI agents, some under evaluation and others wielded by attackers, have accessed sensitive data within Indian organisations. Shockingly, the affected companies did not even detect these breaches themselves; they were notified by the AI model providers who discovered the activity during internal reviews. This underscores a dangerous gap between AI adoption and security readiness.
The Core Problem: Old Gaps, New Speed
The vulnerabilities at the heart of CERT-In's warning are not new. The culprits are familiar security lapses: passwords and API keys accidentally left in public code repositories, outdated test APIs still connected to the internet, and, most critically, weak cloud access controls. Many organisations grant excessively broad permissions to cloud services and user accounts, essentially leaving the digital front door wide open. What has changed is the threat actor. AI-powered tools, whether used by researchers or criminals, can now scan for these weaknesses and exploit them at a speed and scale that far outpaces human capabilities. According to the advisory, these tools have compressed the time from initial breach to final impact from days to mere hours, rendering traditional security response timelines obsolete.
What 'Weak Access Controls' Mean for AI
In the context of AI, weak access controls are particularly dangerous. AI systems are data-hungry, often connecting to vast databases, numerous APIs, and other critical infrastructure. If the roles and permissions governing this access are not strictly defined, the risk of a breach skyrockets. An attacker who compromises a single, 'over-privileged' account could potentially access everything the AI can. This could lead to several nightmare scenarios. One is data poisoning, where an attacker subtly corrupts the training data, causing the AI to produce biased or incorrect results. Another is model theft, where a company's proprietary and expensive AI model is stolen. Attackers could also engage in model inversion, where they reverse-engineer the AI to extract the sensitive private data it was trained on.
CERT-In's Recommendations for a Stronger Defence
In response to these threats, CERT-In has laid out clear, actionable guidance. The focus is on proactive and fundamental security practices. Organisations are urged to meticulously scan all source code, configuration files, and public repositories to find and remove any exposed credentials like passwords or API keys. The advisory stresses the principle of 'least privilege', meaning every user and system should only have the bare minimum access required to perform its function. Unused APIs should be shut down, and administrative panels must be removed from the public internet. Furthermore, CERT-In reiterates the importance of enforcing multi-factor authentication (MFA) across all systems, especially for administrative accounts. Finally, organisations are reminded of their duty to retain security logs for at least 180 days to aid in the investigation of any future incidents.
