CERT-In Technical Guidelines on SBOM, QBOM & CBOM, AIBOM and HBOM (Version 2.0, 9 July 2025) and CERT-In AI security guidance

CERT-In is the Indian Computer Emergency Response Team, the national cyber-security incident-response body within the Ministry of Electronics and Information Technology. Version 2.0 of its Technical Guidelines, dated 9 July 2025, covers the Software Bill of Materials (SBOM), a list of the components in a piece of software, together with Quantum and Cryptographic BOMs (QBOM and CBOM), an AI Bill of Materials (AIBOM) and a Hardware Bill of Materials (HBOM).¹

The guidelines are addressed to entities in the public sector, government and essential services, and to organisations in software export and services, and say departments and organisations 'are encouraged' to make an SBOM a mandatory standard practice in software procurement and development.¹ AI was added because AI systems have their own supply chains of pre-trained models, datasets and frameworks, and a flaw or an undocumented dependency in any of them is a security risk.

CERT-In's first AI-specific document was Advisory CIAD-2023-0015, 'Security implications of AI language based applications', issued on 9 May 2023. It warned that threat actors could use generative AI tools to write malicious code, scan networks, escalate privileges and build malware and ransomware, to generate human-like text for fake news, scams, misinformation, phishing messages and 'deepfake text', and to create fake websites and apps that impersonate AI tools; it also noted that models trained on internet data may collect personal information without consent. Its recommendations included verifying domains and apps claiming to be AI tools, not entering sensitive data, applying content filtering, strengthening monitoring, regular audits, multi-factor authentication and incident-response plans.³

Section 9 of the guidelines defines an AIBOM as a comprehensive list of the components used in building, training and deploying AI models, 'typically' including hardware such as servers, sensors and GPUs, software such as models, frameworks and development tools, and data sources.¹ It identifies four benefits: security (finding vulnerabilities in models and their components and speeding incident response), transparency (documenting algorithms and data sources so stakeholders understand how a system was built), compliance (helping demonstrate adherence to privacy, regulatory and ethical requirements) and risk management (identifying dependencies and failure points).¹

Table 10 sets out the minimum elements of an AIBOM:¹

Section 9.4 contains thirteen recommendations and six best practices. The recommendations say that government, public-sector and essential-services organisations, and organisations involved in AI-driven development and services, should include AIBOM requirements in all AI procurement; that AI solutions supplied to them must be accompanied by an AIBOM containing the specified elements; that they must ensure an AIBOM is kept for all AI systems used, procured and developed; that AIBOMs follow standards such as SPDX or CycloneDX; that suppliers issue VEX documents after a vulnerability or security issue; that consumers build internal AIBOMs from the supplier's and integrate them with vulnerability databases, CERT-In vulnerability notes and threat-intelligence platforms; that AIBOM data be encrypted and access-controlled; and that the AIBOM be updated through the lifecycle. The best practices include keeping a comprehensive inventory, standardised formats, automating AIBOM generation in development pipelines, supporting reproducibility with scripts, weights and configuration, prioritising critical models and tracking model lineage (versions, retraining and modifications).¹

The guidelines are framed as recommendations and technical guidance. Their operative language is that organisations 'should' include AIBOM requirements in procurement and that supplied solutions 'must' be accompanied by one, but they are not rules under the IT Act and carry no penalty. MediaNama described them as guidelines and noted that if the government 'ends up mandating' AIBOMs in procurement, that could set a broader industry standard; it reported no such mandate.² The sources reviewed did not identify any procurement rule, tender condition or scheme that requires an AIBOM.

MediaNama also quoted lawyer Archana Iyer on a practical limit: an AIBOM's benefits depend on what suppliers disclose and on contract terms, for example whether the buyer can tie licensing information in the AIBOM to an indemnity.²

Blueprint of 25 May 2026. CERT-In published the Blueprint for Reducing Exposure and Defending against AI-Assisted Vulnerabilities Exploitation in Digital Infrastructure (Version 1.0, 25 May 2026, 38 pages). It describes AI-assisted threats (rapid reconnaissance, automated vulnerability discovery, personalised phishing, AI-generated malware, 'deepfake-enabled impersonation and fraud' and agentic attack orchestration) and sets out 14 sections of guidance with a phased implementation roadmap. Section 7 says organisations should strengthen software, AI-model and supply-chain visibility through SBOM, AIBOM, QBOM, CBOM and related 'xBOM' mechanisms. Section 12 covers secure adoption and governance of AI: AI usage policies and approval mechanisms; inventories of AI assets and detection of shadow or unauthorised AI; risk assessment before deployment; controls on uploading sensitive information to public AI platforms; prompt-injection defences; data protection; third-party AI provider risk; logging; human oversight and limits on fully autonomous critical actions; AI in software development; model provenance and integrity; and awareness of phishing and deepfakes. Section 9 sets indicative remediation periods, from 12 hours for known exploited flaws on internet-facing systems to five days for high-severity flaws, and the document reminds entities to report cyber incidents to CERT-In within six hours. It names a 'CERT-In AI Cyber Defence Center' as a contact. The Blueprint states that organisations are 'encouraged' to implement it in a risk-informed manner.⁴⁵

AI incident reporting. On 21 September 2026 The Economic Times, as relayed by MediaNama, reported that MeitY sources said India would revise the norms, timeframe and content of AI-related incident reporting. CERT-In's 2022 Directions already require specified cyber incidents to be reported within six hours; MediaNama noted that the report did not say whether AI incidents that are not cyber-attacks would be covered or whether the six-hour period would change, and the India AI Governance Guidelines of November 2025 recommended a voluntary national AI incident database.⁶

Deepfake detection. The Parliamentary Standing Committee on Home Affairs recommended in August 2025 that CERT-In coordinate monitoring and issue deepfake detection alerts, and welcomed C-DAC's Deepfake Detection Tool.⁷ The committee's report is described in a separate entry.

The AIBOM guidelines ask suppliers to document training data provenance, licensing and limitations, which bear directly on copyright and privacy compliance as well as on security. By placing the burden on suppliers to disclose models, datasets and dependencies in a structured, machine-readable form, and by tying the request to public procurement, CERT-In works through the government's purchasing power rather than through regulation.¹

Limits are visible. The guidelines are voluntary and aimed at procurement by government and essential services rather than the private sector at large, and they depend on supplier honesty and capacity to produce accurate AIBOMs for third-party or proprietary foundation models.² They sit within a cyber-security frame: they do not address model behaviour, bias, content harms or intellectual property directly, which are treated by MeitY's IT Rules, advisories and AI Governance Guidelines and by the DPDP Act and copyright law. MediaNama's coverage of the guidelines reported no formal criticism of the AIBOM requirement itself; its report on the 2026 Blueprint described that document as advisory rather than mandatory and noted that no mechanism links it to CERT-In's audit guidelines.²⁵

Version 2.0 of the Technical Guidelines of 9 July 2025 is the operative text. The sources reviewed do not report a later version, a procurement mandate or any enforcement of the AIBOM recommendations as of 15 October 2026. The May 2026 Blueprint reinforces the AIBOM idea for enterprises, and MeitY's reported plan to tighten AI incident reporting, first reported in September 2026, has not been notified in the sources reviewed.⁴⁶