India's CERT-In stands up an 'AI war room' to stress-test frontier models before they flood the market

Verified key facts

• MeitY secretary S Krishnan said on 13 July 2026 that CERT-In has operationalised an AI war room, Business Standard reported.
• The war room works with AI models at roughly 60-70% of the capability of frontier systems from labs such as Anthropic and OpenAI.
• CERT-In has built sandboxes and test environments to identify vulnerabilities and prepare patches ahead of wider Indian access to frontier models.
• Sensitive government workloads are to be processed on-premises, with indigenous models reducing reliance on general-purpose frontier systems.
• An April 2026 CERT-In advisory warned frontier models could automate vulnerability discovery and exploitation.

What the government announced

India's national cyber agency has moved from warning about advanced AI to actively wargaming it. CERT-In has operationalised an artificial intelligence war room, Business Standard reported on 13 July, citing S Krishnan, secretary at the Ministry of Electronics and Information Technology.

The unit works with AI models that reach roughly 60 to 70 percent of the capability of frontier systems, such as those from Anthropic and OpenAI. Inside purpose-built sandboxes, CERT-In teams probe these models for weaknesses, misuse pathways and exploitable behaviour. The goal is preparation, not prohibition.

The 60 to 70 percent figure is telling. Powerful open-weight models now trail the commercial frontier by months rather than years. By testing what is freely downloadable today, CERT-In gets a preview of what paid frontier systems will unleash tomorrow. It is threat forecasting built on publicly available material.

Why a war room, and why now

The trigger sits in CERT-In's own research. An advisory published in April 2026 warned that frontier AI models could demonstrate advanced cyber capabilities, including automated vulnerability discovery and exploitation. In plain terms, the same model that writes your code can find and weaponise flaws in someone else's.

Indian officials have raised this externally too. At G7 discussions this year, India argued that access to frontier AI models is key to fighting AI-enabled cyber threats, Digit reported. You cannot defend against capabilities you have never tested. The war room is the domestic answer to that gap.

This week supplied a live demonstration of the stakes. Microsoft shipped its largest-ever Patch Tuesday, roughly 570 fixes, and credited an AI system with discovering many of the flaws. The same technique works for attackers. Agencies that cannot match that discovery speed will always be reacting to breaches, never preventing them.

How the testing actually works

According to Business Standard, CERT-In has created sandboxes and controlled testing environments where analysts run near-frontier models through attack and defence scenarios. Teams catalogue possible vulnerabilities and prepare mitigation playbooks and patches in advance. When Indian companies and agencies gain wider access to true frontier models, the defensive homework will already exist.

Two design principles stand out. First, sensitive workloads must be processed on-premises within government, not on foreign-hosted APIs. Second, a significant share of indigenous AI development should proceed without heavy dependence on general-purpose frontier models. That reflects a sovereignty calculation as much as a security one.

The approach also differs from what other governments have built. The UK and US house model evaluations in standalone AI safety institutes focused on pre-deployment testing. India has embedded the function inside its incident-response agency instead. That wires model research directly into the team that handles live breaches, advisories and emergency coordination.

The industry context: everyone is bracing

The timing is not accidental. Dataquest India has reported CERT-In warnings through 2026 covering AI-driven threats alongside a heavy run of Microsoft vulnerabilities affecting Indian organisations. Defensive teams are already stretched, and AI promises to accelerate the attacker side first.

Banks are a particular worry. Indian financial institutions face growing threats from frontier-model-assisted attacks, according to industry reporting on the CERT-In initiative. Phishing that adapts in real time, deepfake-driven fraud and automated reconnaissance all get cheaper as model capability spreads.

The war room also gives India sharper standing in global AI negotiations. A government that publishes its own adversarial findings can push labs for safeguards with evidence, not anxieties. That matters as frontier vendors court Indian enterprises, and as New Delhi weighs access rules for advanced models. It also builds the technical cadre India has lacked in cyber diplomacy.

What it means for Indian businesses and users

For enterprises, the war room signals coming compliance expectations. CERT-In already mandates breach reporting timelines. It is reasonable to expect AI-specific advisories, model-testing guidance and possibly disclosure norms for AI-related incidents to follow from this work.

• Security teams should assume AI-assisted attacks in threat models from this year onward.
• Firms deploying large models should sandbox and red-team them before production, mirroring CERT-In's approach.
• Government suppliers should prepare for on-premises processing requirements on sensitive data.
• Startups building on frontier APIs should track MeitY guidance on indigenous alternatives.

For ordinary users, the benefit is indirect but real. A national agency that understands how models fail is better placed to protect UPI rails, telecom networks and government databases from AI-enabled attacks. India is unusual among large economies in operationalising this capability inside its emergency response team.

What to watch next

Three markers will show whether the war room has teeth. Watch for CERT-In technical advisories that cite its own model-testing findings. Watch for procurement rules steering ministries toward vetted, domestically hosted models. And watch whether the IndiaAI Mission's compute buildout links funding to security evaluations.

The deeper question is strategic. India is betting it can capture AI's economic upside while containing its offensive potential, without waiting for global governance frameworks to mature. The war room turns that bet into an institution. Its first public advisory will tell us how the experiment is going.

Sources

• Business Standard - CERT-In operationalises AI war room to work on frontier AI models (13 July 2026)
• Dataquest India - CERT-In warning 2026: AI cyber threats and Microsoft vulnerabilities in India (July 2026)
• Digit - India at G7 2026: access to frontier AI models key to fight cyber threats (2026)