In response to growing concerns about the cybersecurity risks posed by AI models, the Indian Computer Emergency Response Team (CERT-In) has unveiled a new framework for organisations to protect against AI-assisted cyber attacks through faster detection, shorter incident response timelines, better reporting, and more.
The new 38-page blueprint released by India’s nodal cybersecurity agency on Monday, May 25, raises the alarm on collapsed attack timelines due to recent advances in AI.
The document titled ‘Blueprint for Reducing Exposure and Defending against AI-Assisted Vulnerabilities Exploitation in Digital Infrastructure’ does not explicitly mention Anthropic’s unreleased Claude Mythos AI model, which is said to be exceptional at scanning software systems and identifying previously unknown (zero-day) vulnerabilities in limited timeframes.
However, CERT-In has advised organisations to patch known vulnerabilities affecting internet-facing and ‘crown-jewel’ systems within 12 hours of being flagged where feasible to protect against malicious attacks.
The agency has also recommended critical externally exposed vulnerabilities to be patched or mitigated within a day, along with vulnerabilities affecting internal systems provided other controls are already in place. In case of high-severity vulnerabilities, organisations can take up to five days while vulnerabilities in high-value systems should be resolved within three days, as per the blueprint.
CERT-In’s blueprint comes as governments and organisations around the world scramble to prepare for how attackers might eventually use AI to carry out cyber attacks. Last month, Finance Minister Nirmala Sitharaman chaired a high-level meeting over concerns that Mythos could pose significant risks to India’s banking sector. The Indian government is also in conversation with Anthropic’s senior leadership in the US on the issue, The Indian Express had previously reported.
“The rapid advancement and accessibility of artificial intelligence (AI), including generative AI, large language models (LLMs), autonomous agents, and AI-enabled automation platforms, are significantly transforming the cybersecurity landscape,” CERT-In said in the blueprint.
“Threat actors are increasingly leveraging AI capabilities to accelerate reconnaissance, automate vulnerability discovery, generate highly targeted phishing campaigns, develop adaptive malware, and enhance the scale and speed of cyber-attacks,” it added.
The agency also reiterated the requirement for entities to report cyber incidents within six hours. However, the rest of the blueprint does not impose any legal obligations on entities, and only offers recommendations to help organisations strengthen their resilience against AI-enabled cyber threats.
Here are the key concerns, major threat areas, and recommended pre-incident and post-incident measures.
Given the evolving landscape of AI-assisted cyber threats, CERT-In said that organisations should take note that timelines for exploiting vulnerabilities have reduced significantly owing to advanced AI models and tools, to a point where even periodic audits and reactive responses may no longer be enough.
CERT-In also emphasised that , and that organisations need to shift away from static cyber defence models. Additionally, the rise of autonomous AI agents poses the threat of “semi-autonomous or fully automated cyber operations capable of accelerating multiple stages of the cyber kill chain, including reconnaissance, exploitation, privilege escalation, lateral movement, and data exfiltration.”
As organisations increasingly depend on cloud-native infrastructure, APIs, interconnected digital services, software supply chains, operational technologies, and AI-enabled platforms, have led to an expansion of the organisational attack surface, according to CERT-In. AI-assisted cyber attacks generally take the form of:
-Rapid reconnaissance and attack surface mapping.
-Automated vulnerability discovery and exploit development.
-Highly personalised phishing and social engineering campaigns: This includes spear phishing campaigns, executive impersonation, deepfake voice and video fraud, business email compromise, credential theft campaigns, AI-generated social engineering at scale.
-AI-generated malware and malicious scripting: AI-assisted offensive tooling may be used for End-to-end Cyber Kill Chain (CKC) execution, Malware modification and obfuscation, Adaptive payload generation, Automated scripting, Evasion of static detection controls, Semi-autonomous attack execution, Lowering of technical entry barriers, enabling even untrained threat actors to launch sophisticated cyber-attacks at scale
-Deepfake-enabled impersonation and fraud.
-Automated attack orchestration.
-Adaptive evasion techniques.
“Organisations deploying AI-enabled systems may themselves become targets of adversarial attacks against AI models, inference systems, retrieval mechanisms, and AI-integrated workflows,” CERT-In said, highlighting prompt injection attacks, model manipulation, training data poisoning, insecure AI integrations, and AI model theft as potential risks.
It further cautioned against the unrestricted use of public AI platforms.
CERT-In has urged organisations to adopt AI-enabled adaptive practices, effectively encouraging the use of AI tools to counter AI-driven threats. It also pointed out that perimeter-centric and periodic compliance-driven security approaches, while required, are not sufficient against malicious, AI-enabled attacks.
As precautionary measures, CERT-In said organisations should assume breach to prepare for rapid detection, containment, and recovery from compromise scenarios. It also recommended adopting the principle of zero trust security and setting up multi-factor authentication (MFA), privileged access management (PAM), micro segmentation, conditional access, and session monitoring.
To protect against software supply chain attacks, CERT-In recommended the adoption of Software Bill of Materials (SBOM), AI Bill of Materials (AIBOM), Quantum Bill of Materials (QBOM), Cryptographic Bill of Materials (CBOM), and related xBOM mechanisms to help improve transparency, component visibility, dependency tracking, provenance validation, vulnerability, impact assessment, rapid exposure identification, and coordinated remediation
CERT-In further proposed that organisations undertake continuous audits by the agency’s empanelled auditors. “Organisations should conduct Red Teaming & cybersecurity audits, security assessments, adversarial simulations, and resilience validation exercises to assess effectiveness of implemented controls and operational preparedness,” it said.
“Where applicable, such assessments may be conducted through CERT-In empanelled Information Security Auditing Organisations in alignment with the Comprehensive Cyber Security Audit Policy Guidelines and other relevant guidelines issued by CERT-In from time to time,” the document read.
