AI security is becoming a board-level issue in India: What CISOs and business leaders need to change

AI has become a force multiplier. Boards no longer debate whether to integrate this transformative technology into their everyday workflow, but how to do so at scale without increasing the risk surface.

Indian boards are no exception when it comes to AI adoption. However, they are approving many AI decisions that they cannot anticipate. The financial cost of that blind spot is rising with every reporting cycle. According to a 2026 breach cost study, the average data breach in India now costs ₹255 million. The figure is up 15.9 percent from the ₹220 million recorded one year earlier. Today, AI security has become a question of trust, regulation, and long-term enterprise value.

This guide explains board-level AI risk for Indian leaders. It shows what CISOs and business leaders must change over the next twelve months.

Why AI security has moved from the server room to the boardroom in India

AI now touches revenue, customer data, and regulatory standing in most large Indian firms. Therefore, directors can no longer treat AI security as a back-office task.

From IT problem to enterprise risk

Earlier, security teams managed most risk inside the corporate network and its servers. AI has expanded that risk across business functions, as models access customer records, draft contracts, and trigger transactions. The financial exposure is growing. India's average data breach cost reached ₹255 million in 2026, a 15.9% increase over 2025. With AI influencing financial, operational, and customer-facing decisions, a single security failure can now affect the entire enterprise.

What Indian boards and CEOs are saying about AI and cyber risk

A 2026 survey of 138 Indian businesses ranked AI as the top cyber budget priority at 46 percent. Cloud security followed at 33 percent, and managed services reached 28 percent. Meanwhile, a 2026 board agenda warns that directors often receive fragmented views of AI risk.

Why 2026 feels different from earlier cyber cycles

Three major events arrived within twelve months of each other in India. The government published the India AI Governance Guidelines on November 5, 2025. The DPDP Rules followed on November 13, 2025, with a phased timeline. Then CERT-In released its blueprint on defending against AI-assisted attacks on May 25, 2026.

The trust question behind every AI deployment

Customers share personal data because they trust the firm to protect it. AI multiplies the number of places where that data travels and rests. Therefore, every new AI deployment raises a critical question for the full board: Can the company prove that its AI systems behave safely and lawfully today?

Trust depends on understanding the threats.

We will now examine the threats that put this trust at risk across Indian enterprises.

The AI threat landscape facing Indian enterprises

The AI threat landscape India faces now combines faster attacks with new weaknesses inside enterprise AI. Below, we discuss the challenges facing Indian enterprises.

Shadow AI and unsanctioned tools

Employees put sensitive data into AI tools for better work efficiency. Shadow AI risk grows when security teams never approve the tools that employees use. Shadow AI added ₹17.9 million to the average breach cost in India in 2026. Indian Computer Emergency Response Team guideline advises organizations to maintain AI inventories and monitor unauthorized AI usage.

Attacks on AI systems: Prompt injection, data poisoning, and model theft

Attackers also target the AI systems that companies build, buy, and connect to data. Prompt injection tops the risk list that the OWASP GenAI Security Project maintains for LLM applications. Data poisoning corrupts training inputs, and model theft copies proprietary models without permission. Therefore, boards should treat every model as an asset that needs protection.

AI-powered attacks: Phishing, deepfakes, and automated reconnaissance

Phishing was the most common initial attack vector in India at 19 percent. That category includes voice and SMS phishing, and drive-by compromise followed at 16 percent. Furthermore, 26 percent of malicious breaches in India were AI-generated. CERT-In adds that AI enables executive impersonation and business email compromise at scale.

Agentic AI and non-human identities

AI agents act with some independence, using credentials, APIs, and tool connections. Each agent is a non-human identity that needs an owner, limits, and an audit trail. It is worth highlighting that CERT-In’s May 2026 blueprint also calls for clear governance of agentic AI systems.

AI-powered data attacks and DPDP Act exposure

AI-powered attacks can expose personal data by exploiting weak access controls, manipulating AI systems into retrieving sensitive records, or extracting information from model outputs. Under India’s DPDP Act, organizations must implement reasonable security safeguards to prevent personal data breaches. A failure can attract penalties of up to ₹250 crore under Section 8(5) read with Section 33 and the Schedule. Therefore, enterprises must monitor how AI systems access, process, and share personal data, while maintaining evidence to support breach investigation and regulatory reporting.

The financial case: What a breach costs Indian organizations now

Data breach cost India figures give boards the clearest financial evidence for action. The numbers below come from IBM’s Cost of Data Breach Report, 2026.

Rising average breach costs and record exposure

The average breach in India cost ₹255 million in 2026, a record high. Each breach also exposed 39,500 records on average, up from 38,200 a year earlier. Financial services recorded the highest cost at ₹409 million, followed by technology at ₹357 million.

The price of shadow AI and weak governance

Shadow AI, regulatory non-compliance, and cloud migration were the top three cost drivers in India. Shadow AI alone added ₹17.9 million to the cost of a breach where it was present. Therefore, weak governance carries a direct price, not only a compliance risk.

How AI-powered attacks can lead to DPDP penalties

AI-powered attacks can expose personal data by exploiting vulnerabilities in AI systems, manipulating models through prompt injection, or gaining unauthorized access to connected databases and applications. Under the DPDP Act, organizations must implement reasonable security safeguards to prevent personal data breaches. If an AI-related attack exposes personal data because these safeguards are inadequate, the Data Fiduciary may face a penalty of up to ₹250 crore. Failure to notify the Data Protection Board and affected individuals as required can attract an additional penalty of up to ₹200 crore. The combined penalty ceilings can therefore reach ₹450 crore for a single incident. The actual penalty depends on the circumstances and the Board’s determination.

Operational, reputational, and litigation costs boards often miss

Breach reports rarely capture every cost that a board will eventually face. Consider these categories:

• Business interruption while teams restore systems and verify the integrity of data.
• Customer attrition after a public disclosure of a breach involving personal data.
• Legal fees and claims from customers, partners, and regulators after an incident.
• Higher insurance premiums and tougher terms at the next annual policy renewal.

How AI and automation in defense reduce loss

AI-powered security operations can reduce the time required to detect threats, investigate suspicious activity, and contain breaches. This helps limit operational disruption, reduce the likelihood of prolonged customer data exposure, and provide evidence for legal and regulatory investigations. AI tools can also help discover and classify personal data, monitor access, and maintain audit trails that support DPDP compliance. Organizations without AI and automation in security operations faced an average breach cost of ₹316 million in India. Offensive security practices, including red teaming and penetration testing, reduced breach costs by ₹24.7 million on average.

India’s regulatory pressure points boards cannot ignore

India regulates AI through existing laws and sector rules, so AI security obligations come from several regulators. Boards therefore face overlapping expectations, and each one carries consequences.

India AI governance guidelines and the seven sutras

The Ministry of Electronics and Information Technology released the guidelines on November 5, 2025. The framework is voluntary, and it rests on seven guiding principles called sutras. Those sutras adapt the principles that the RBI published for the financial sector in August 2025. The guidelines also favor existing laws and sector regulators over a new AI statute.

DPDP Act and rules 2025: safeguards, 72-hour breach reporting, and penalties

The government notified the DPDP Rules on November 13, 2025, with a phased rollout. Full obligations on notice, security safeguards, and breach reporting apply from May 13, 2027. A Data Fiduciary must report a breach to the Data Protection Board within 72 hours of awareness. Penalties reach ₹250 crore for weak safeguards and ₹200 crore for missed notification. DPDP Act AI compliance therefore begins with knowing where personal data enters each model.

RBI FREE-AI framework and board-approved AI policy

The RBI committee published its FREE-AI report, with 26 recommendations, on August 13, 2025. The RBI FREE-AI framework asks regulated entities to adopt a board-approved AI policy. That policy should cover governance, risk appetite, auditability, consumer protection, and model lifecycle controls.

SEBI proposals on responsible AI and ML

SEBI issued a consultation paper on responsible AI and ML usage on June 20, 2025. It proposes that senior managers with relevant technical knowledge must oversee every AI model. The paper also asks market participants to report technical glitches and data breaches. These SEBI AI guidelines began as proposals, so firms should confirm their status.

CERT-In guidelines: AIBOM, audit policy, and the 2026 AI defense blueprint

CERT-In AI security guidelines now ask for more than periodic checks. The May 25, 2026 blueprint tells organizations to maintain an AI Bill of Materials, known as AIBOM. It also encourages continuous audits and assessments through auditors that CERT-In has empaneled. It further sets a 12-hour patch target for known exploited flaws on internet-facing systems, where feasible. Legal analysts describe the recommendations as advisory, yet they signal clear regulatory direction.

How these frameworks overlap and where gaps remain

Gaps remain because no single regulator owns enterprise-wide AI risk. Agentic AI, model theft, and AI incident reporting lack uniform rules across sectors. Therefore, a firm must build one control set that satisfies the strictest applicable rule.

Rules describe what regulators expect, and the next section shows how far practice lags.

The readiness gap between boards, CISOs, and AI adoption

Ambition on AI adoption has outrun the controls that should protect it. The data below shows where the largest gaps sit in Indian organizations.

Confidence in defenses versus control over AI use

Many leaders feel confident about traditional perimeter defenses and network monitoring tools. Control over how employees and business applications use AI is far weaker. Nearly 60 percent of organizations in India lacked a finished AI governance policy in 2025.

Missing AI access controls and governance policies

Only 37 percent of Indian organizations reported AI access controls in the 2025 study. Only 42 percent had policies to manage AI or detect shadow AI. Without access controls, any employee or agent can reach sensitive models and data.

Limited visibility into AI agents and tools

Boards cannot govern AI tools and agents that security teams have never inventoried. CERT-In now advises organizations to keep AI inventories and monitor unauthorized AI use. Visibility is therefore the first control that every other control depends on.

Skills and resource gaps in security teams

Skills shortages slow both AI adoption and AI defense, according to the 2026 survey of Indian businesses. Moreover, 73 percent of Indian organizations in the 2025 study reported limited or no use of AI-driven security tools. Training programs and managed services can close part of this gap quickly.

Boards asking for cyber expertise at director level

Boards increasingly want directors who can question cyber and AI risk with confidence. A 2026 board agenda notes that digital dependence is scaling faster than institutional capacity. Therefore, director-level cyber expertise has moved from a preference to a priority.

Closing these gaps starts with clear ownership, so the next section defines what boards must own.

What board-level ownership of AI security means

Board-level AI risk oversight means that the board owns direction, while management owns daily execution. Confusing these two roles creates gaps that attackers and regulators both find.

Oversight versus management: drawing the line

Board cybersecurity oversight in India now covers AI systems, not only networks and servers. The board approves AI strategy, risk appetite, and policy, then reviews results regularly. Management designs the controls, runs operations, and escalates incidents without delay. Directors ask hard questions about AI risk, but they do not configure any tools.

Which committee owns AI risk

Most boards assign AI risk to the audit committee or the risk management committee. Some create a technology committee, especially in financial services and large technology firms. Whichever model a board chooses, it should name one committee as the accountable owner. Shared ownership without a named lead usually means no ownership at all.

Directors’ accountability and duty of care

Directors owe a duty of care to the company and its shareholders. That duty now reasonably includes informed oversight of cyber and AI risk. Directors should confirm the exact scope of this duty with qualified legal counsel. Moreover, DPDP penalties of up to ₹250 crore make weak oversight very expensive.

Approving an AI policy and risk appetite

The RBI FREE-AI report asks regulated entities for a board-approved AI policy. Every board, whether regulated or not, can adopt the same sound practice. The policy should state which AI uses the company accepts, restricts, and prohibits. Moreover, it should define the loss limits that trigger escalation to the full board.

Building AI and cyber literacy in the boardroom

The FREE-AI report recommends training programs for board members, senior management, and staff. Short quarterly briefings with real incident case studies work better than annual lectures. Directors do not need engineering skills, but AI security for boards demands enough fluency to challenge answers.

Boards set the direction, but CISOs must change how they work, and the next section explains how.

What CISOs need to change

CISO AI governance expectations in India are shifting the role from control owner to trusted business risk advisor. Six practical changes define that shift for CISOs in India’s fast-moving AI environment.

Move from gatekeeper to business enabler

Blocking every AI tool simply pushes employees toward shadow AI that nobody monitors. A good CISO offers approved tools, clear rules, and fast approval for new use cases. Speed with safeguards earns the CISO a trusted seat in every business decision.

Own the AI risk register and inventory

The CISO should maintain a live inventory of models, agents, datasets, and vendors. Each entry needs a named business owner, a risk tier, and a review date. This register then becomes the single source of truth for all board reporting.

Translate technical risk into business impact

Boards respond to rupees, downtime, and regulatory exposure, not to raw vulnerability counts. For example, estimate how many records one exposed AI agent could reach, then compare that with the 39,500 average.

Build board communication skills

Effective board updates run ten minutes, lead with decisions, and avoid technical terms. Use one page, one trend chart, and one clear request for each meeting. Practice the presentation with a trusted director before the formal board meeting begins.

Shift from annual reviews to continuous assurance

Annual reviews are insufficient for AI systems that evolve continuously through changes to models, data, applications, and integrations. Mature AI security and governance require continuous assurance, with ongoing monitoring of system behavior, access controls, vulnerabilities, and compliance. CERT-In’s blueprint emphasizes continuous monitoring, vulnerability scanning, and audits rather than relying solely on periodic assessments. CISOs should automate evidence collection, track control effectiveness, and investigate exceptions throughout the year. This helps ensure that security and governance controls remain effective as AI systems change.

Close the AI skills gap inside the team

Train security analysts in model risk, prompt injection, data poisoning, and agent permissions. Pair security analysts with data science teams so both groups share one vocabulary. Where hiring is slow, a managed security partner can fill urgent gaps.

However, CISOs cannot succeed alone, so the next section shows what business leaders must change.

What business leaders need to change

AI security succeeds only when every executive owns a defined part of it. Each leader named below carries a distinct and measurable responsibility for AI risk.

CEOs and CFOs: fund AI security as a growth enabler

CEOs and CFOs should treat AI security funding as a growth enabler, not overhead. Firms that used AI and automation extensively in security saved money and responded faster. The ₹316 million average cost without such tools makes the business case clear.

CIOs, CTOs, and data officers: share accountability

These leaders build and run the AI systems that CISOs must protect. They should share accountability for data quality, model changes, and platform configuration. Joint ownership prevents the common pattern where each team assumes another team handles risk.

Legal, compliance, and privacy: align on regulation

These teams translate DPDP, RBI, SEBI, and CERT-In expectations into internal requirements. They should maintain one regulatory map that links each rule to a named control. Moreover, they should review every AI vendor contract for data use and notification terms.

HR and procurement: policy, training, and vendor terms

HR owns acceptable use training and the disciplinary steps for repeated policy violations. Procurement owns the vendor terms, including audit rights and incident notification duties. Both functions turn written policy into daily behavior across every department in the company.

Business unit heads: name an owner for every AI use case

Every AI use case in a business unit needs one named business owner. That owner approves the purpose, monitors outcomes, and reports incidents to the CISO. Unowned use cases are where shadow AI and unmanaged risk usually begin.

With roles now clear, the company can build its AI governance foundation, which comes next.

Building an AI governance foundation

AI governance India programs need a foundation that turns written principles into repeatable daily practice. Five building blocks form this foundation, and each one supports the next.

Create a complete AI inventory

List every model, agent, API, dataset, and embedded AI feature in use. Include tools that employees adopted without approval, since shadow AI hides in this group. Refresh the inventory monthly, because AI usage changes faster than annual audits can track.

Classify use cases by risk tier

Sort every use case into high, medium, and low risk tiers based on impact. High-risk cases involve personal data, financial decisions, customer rights, or critical operations. The assigned tier then determines testing depth, approval level, and monitoring frequency.

Write an acceptable use policy people will follow

An acceptable use policy fails when employees cannot understand or remember its rules. Use plain language, short examples, and a clear list of approved tools. Moreover, state what data types employees must never enter into public AI tools.

Define roles, decision rights, and escalation paths

Define who approves, who builds, who monitors, and who can stop an AI system. Document escalation paths with named roles, contact details, and expected response times. Clear decision rights prevent costly delays when a live incident demands fast action.

Map controls to NIST AI RMF and ISO 42001

The NIST AI risk management framework organizes work into govern, map, measure, and manage functions. ISO/IEC 42001 provides a certifiable management system standard for organizations that build or use AI. Mapping Indian requirements to both gives multinational firms one consistent control set.

Securing the AI lifecycle with technical controls

Securing AI systems requires controls that address vulnerabilities across the AI lifecycle, from data ingestion and model development to deployment and inference. The OWASP Top 10 for LLM Applications identifies key application-level risks, while MITRE ATLAS provides a framework for understanding adversarial tactics and techniques targeting AI systems. Together, they help CISOs identify threats, map security controls to specific attack scenarios, and strengthen protection for AI models, data, identities, and outputs.

Data security posture and classification

AI systems inherit weaknesses in the personal data they access and process. Under India’s DPDP Act, Data Fiduciaries must implement reasonable security safeguards to prevent personal data breaches. Enterprises should discover and classify personal data across cloud and on-premises environments before connecting these sources to AI systems. Classification should guide access permissions, data masking, minimization, and retention policies. Organizations should also maintain visibility into how AI models and applications access and process personal data to support compliance and breach investigation.

Model, pipeline, and training data protection

Protect training data from tampering, because poisoned inputs corrupt model behavior silently. Restrict access to model weights, and sign every model artifact before deployment. Secure the pipeline that moves data and models from development into production.

Identity and access management for AI systems

Treat each model, agent, and service account as an identity with defined limits. CERT-In recommends multi-factor authentication, privileged access management, and continuous verification under zero trust principles. Remove unused credentials quickly, because stale tokens give attackers silent and lasting access.

Runtime protection against prompt injection and misuse

Prompt injection tricks a model into ignoring its instructions and following an attacker’s text. OWASP acknowledges that no foolproof prevention exists, so layered controls become essential. Filter inputs and outputs, limit tool permissions, and monitor for unusual model behavior.

Reviewing AI-generated code and establishing code provenance

AI coding assistants can introduce insecure code patterns, vulnerable dependencies, and components with unclear origins into production software. Enterprises should maintain code provenance by tracking which code was generated or modified by AI, the tools and models involved, and the sources of third-party dependencies. CERT-In recommends static testing, dynamic testing, and dependency analysis for AI-generated code. It requires developers to document AI contributions, verify dependencies, and conduct human code reviews before approving changes for production. This creates traceability and helps security teams investigate vulnerabilities and establish accountability.

Logging and monitoring for audit readiness

Log prompts, outputs, tool calls, and data access for every high-risk AI system. These records support incident investigation, and they give auditors direct proof of compliance. Moreover, DPDP breach rules also expect organizations to keep evidence logs.

Controlling shadow AI without slowing the business

Outright bans usually fail because employees choose speed over policy every time. A better approach offers safe options and then watches for the exceptions.

Discover what employees already use

Start with network logs, browser telemetry, and expense records to find AI tools in use. Then survey teams directly, because honest answers reveal use cases that logs miss. Treat the first discovery exercise as a baseline, not as a final answer.

Set clear guardrails for sensitive data

Define which data classes employees may never enter into external AI tools. Examples include customer personal data, source code, unreleased financials, and legal documents. Technical controls, such as data loss prevention, should enforce these rules automatically.

Offer approved AI tools as safe alternatives

Employees use shadow AI because it solves real work problems quickly and easily. Provide approved enterprise AI tools that include contractual data protection and single sign-on access. Safe alternatives reduce risky behavior far more effectively than warnings and reminders ever will.

Monitor continuously

CERT-In’s blueprint asks organizations to monitor and identify shadow or unauthorized AI usage. Run discovery scans every month and review the exceptions in the risk committee. Annual audits cannot match the pace at which new AI tools appear online.

Shadow AI involves people, whereas agentic AI involves software that acts alone, which the next section addresses.

Securing agentic AI and non-human identities

Agents plan tasks, call tools, and change systems with limited human supervision. Therefore, agentic AI security requires stricter controls than ordinary software ever needed.

Scope permissions and apply least privilege

Give each agent only the access that its specific assigned task requires. Use short-lived credentials and separate identities for each agent running in production. OWASP lists excessive agency among the top ten risks for LLM applications.

Keep humans in the loop for high-impact actions

Require human approval for actions that move money, delete data, or change access rights. Define clear thresholds, such as transaction value or data volume, that trigger mandatory review. SEBI’s proposals similarly call for human oversight to prevent over-reliance on AI.

Govern tool connections and integrations

Every connection between an AI agent and an enterprise system expands the attack surface. An insecure MCP server, API, or skill can expose sensitive data or allow an agent to perform unauthorized actions. Maintain an approved inventory of tools, APIs, MCP servers, and data sources available to each agent. Apply least-privilege permissions, validate access requirements, and monitor tool activity. Review new integrations through the same security assessment and change-management process used for production software. Reassess existing connections regularly to ensure they remain secure and necessary.

Maintain auditable agent activity trails

Record every agent decision, tool call, and data access with a timestamp. Store these logs in a protected location where the agent itself cannot alter them. Auditable trails let investigators reconstruct an incident within the 72-hour DPDP window.

Agents also depend on outside vendors, models, and platforms, so the next section covers supply chain risk.

Managing third-party and AI supply chain risk

Third-party AI risk grows with every new vendor and every embedded AI feature. Boards should demand clear visibility into that exposure before approving further expansion.

Due diligence for AI vendors and embedded AI

Ask vendors where they host models, which data trains them, and who can access prompts. Review security certifications, incident history, and sub-processor lists before signing any agreement. Include AI features inside existing software, since these often arrive through quiet updates.

Using AIBOM to see what you run

An AI Bill of Materials lists the models, datasets, and components inside an AI system. CERT-In’s blueprint recommends AIBOM alongside software, quantum, and cryptographic bills of materials. Ask vendors to supply an AIBOM so your team can trace inherited risk.

Contract terms: audit rights and incident notification

Contracts should grant audit rights over AI security controls and data handling practices. They should also require incident notification fast enough to meet the 72-hour DPDP window. CERT-In advises vendor assessments, contractual controls, dependency visibility, and regular supplier reassessment.

Concentration risk in cloud and model providers

Many Indian firms rely on the same few cloud and model providers. One outage or breach at a single provider can disrupt many companies at once. Therefore, boards should ask for exit plans and fallback options for critical AI services.

Even strong supplier controls cannot prevent every failure, so the next section prepares you for AI incidents.

Preparing for AI-specific incidents

An AI incident response plan needs the same rigor as a data breach plan. It also needs steps that address model behavior, data exposure, and agent actions.

Updating incident response playbooks for AI failures

Add scenarios for prompt injection, model poisoning, data leakage through AI tools, and agent misuse. Assign owners, containment steps, and communication leads for each scenario in the playbook. Test each playbook with realistic AI scenarios at least twice each year. Gruve’s work on [DFIR for enterprise AI incidents] shows how investigators approach these cases.

Meeting the 72-hour DPDP notification window

Under Rule 7 of the DPDP Rules, a Data Fiduciary must inform affected individuals without delay. It must then submit a detailed report to the Data Protection Board within 72 hours. Design the process for 72 hours, even though the Board may grant extensions in rare cases. DPDP reporting does not replace the separate reporting duty that applies under CERT-In directions.

Running board-level tabletop exercises

Run a board-level tabletop exercise at least once every year with realistic AI scenarios. A good scenario might involve an AI agent that leaks customer data to an outside party. Directors practice decisions on disclosure, regulator contact, and customer communication under time pressure.

Reporting AI incidents to regulators and stakeholders

The RBI FREE-AI report includes an indicative AI incident reporting form for regulated entities. SEBI proposes that market participants report technical glitches and data breaches to the regulator. Prepare templates and approval chains now, so reporting does not depend on improvisation.

Preparation produces evidence, and boards need that evidence in a clear format, which the next section covers.

Measuring and reporting AI security to the board

Directors cannot oversee any risk that the organization fails to measure consistently. The right metrics turn AI security into a standing agenda item at every meeting.

Metrics that matter: coverage, exposure, and response time

Three metric groups give directors a balanced and honest view of progress:

• Coverage: the percentage of AI systems that appear in the inventory with a named owner.
• Exposure: the number of high-risk AI use cases without completed testing or approved controls.
• Response: the median time to detect, contain, and report an AI-related incident.

Building a quarterly AI risk dashboard

Build one dashboard that refreshes every quarter and fits on a single page. Show trends against the previous quarter, because isolated numbers hide the real direction. Add a short list of decisions that the board must make this quarter.

Quantifying AI cyber risk in rupees

Half of Indian respondents in the 2026 survey use cyber risk quantification to a significant extent. AI cyber risk quantification estimates the probable loss from each major AI scenario. Use breach cost benchmarks, such as ₹255 million, to calibrate those estimates.

Ten questions every director should ask

• Which AI systems and agents does the company run today, and who owns each one?
• Which high-risk use cases process personal data under the DPDP Act?
• What does our board-approved AI policy permit, restrict, and prohibit?
• How do we detect shadow AI, and how often do we scan?
• Which vendors embed AI in the services that we depend on?
• Can we meet the 72-hour DPDP notification window today?
• When did we last run a board-level AI incident exercise?
• What is our estimated rupee loss from the top three AI scenarios?
• Which controls map to RBI, SEBI, CERT-In, and DPDP expectations?
• Do we have enough skilled people, or should a partner supplement them?

Metrics and risks vary by industry, so the next section outlines priorities for each major sector.

Sector-specific priorities across India

Every sector faces the same core risks, but each faces a different regulatory mix. The priorities below reflect those differences in data, regulation, and operating models.

Banking, financial services, and insurance

Financial services recorded the highest breach cost in India at ₹409 million. The RBI’s FREE-AI report and SEBI’s AI and ML paper apply directly to this sector. Priorities include a board-approved AI policy, model audits, and consumer protection safeguards.

Healthcare and life sciences

These organizations process highly sensitive personal data, which raises their DPDP penalty exposure considerably. Prioritize strict access controls on clinical AI tools, data minimization, and thorough vendor assessments.

IT services and global capability centers

Global capability centers and IT service firms handle client data under several legal regimes simultaneously. Technology firms recorded an average breach cost of ₹357 million in 2026. Priorities include client contract alignment, secure AI coding practices, and strict workspace isolation.

Manufacturing and critical infrastructure

AI now supports predictive maintenance, quality control, and supply planning in factories. A compromised model can halt production lines or create physical safety incidents. Therefore, prioritize segmentation between operational technology and AI systems, plus continuous monitoring.

Startups and mid-sized enterprises

Smaller firms lack large security teams, yet they face the same DPDP obligations. Start with an AI inventory, an acceptable use policy, and approved tools. Managed security services give mid-sized enterprises expert coverage without a large fixed cost.

Sector priorities need a realistic timeline, so the next section provides a twelve-month roadmap.

A practical roadmap for the next 12 months

A roadmap converts good intent into dated, measurable commitments that boards can track. This AI security roadmap spans twelve months, and it aligns with the May 2027 DPDP deadline.

First 30 days: inventory and ownership

Appoint an executive owner for AI security and name a committee sponsor. Complete the first AI inventory, including shadow AI discovered through logs and surveys. Present the baseline findings to the full board along with the top five exposures.

By 90 days: policy, controls, and board reporting

Approve the AI policy, the risk appetite statement, and the acceptable use rules. Deploy core controls on high-risk systems, including access management, logging, and data classification. Launch the quarterly AI risk dashboard so the board receives its first formal report.

By six months: testing, red teaming, and vendor reviews

Complete red teaming on the highest-risk AI systems and fix all critical findings. Review the top vendors against the new contract terms and AIBOM requirements. Run the first board-level tabletop exercise using a realistic AI incident scenario.

By 12 months: continuous monitoring and audit readiness

Operate continuous monitoring across AI systems, agents, data stores, and third-party connections. Prepare audit evidence, including logs, policies, test reports, and board minutes, for any regulator. Repeat the inventory and risk tiering to capture new AI use cases.

Aligning the roadmap with the May 2027 DPDP deadline

The DPDP Rules require full compliance with notice, safeguards, and breach reporting from May 13, 2027, backed by frameworks like ISO 42001 (AI Management System) and NIST AI RMF (Govern, Map, Measure, Manage) to ensure secure, risk-aware data governance. MeitY has discussed shortening timelines for some entities, but no final change was confirmed. Therefore, plan for earlier readiness and confirm the current dates with counsel. A good plan still fails when the company repeats predictable errors, which the next section lists.

Common mistakes to avoid

Five mistakes recur across Indian enterprises that adopt AI at considerable speed. Each one is avoidable when the board pays close attention at an early stage.

Treating AI security as a tool purchase

No single product delivers governance, because governance depends on people, process, and ownership. Buy tools only after the inventory, policy, and risk tiers are defined.

Writing policies without enforcement

A policy that nobody monitors becomes mere decoration within a few short months. Pair every rule with a technical control, a measurable check, and a consequence.

Leaving AI risk with IT alone

IT teams cannot decide business risk appetite, legal exposure, or customer commitments. Therefore, assign cross-functional ownership with clear board-level sponsorship and named business owners.

Ignoring embedded AI in existing software

Software vendors add AI features through routine updates that nobody reviews carefully. Review existing contracts and request AIBOM disclosures for every business-critical application today.

Reporting activity instead of risk reduction

Counting completed trainings and deployed tools does not show whether risk has actually fallen. Report measured changes in exposure, response time, and estimated loss instead of activity counts.

Avoiding these mistakes is easier with experienced support, so the next section explains partner value.

How the right security partner supports board-level AI governance

Gruve is an AI and cybersecurity services company that builds and secures AI systems. That combination matters because AI security needs both engineering depth and security operations discipline. Boards gain three benefits:

• One partner covers AI engineering, SOC operations, data security, and red teaming.
• Controls map to Indian regulation and to global frameworks such as NIST AI RMF and ISO 42001.
• Evidence arrives in board-ready formats that regulators and auditors can verify quickly.

Managed detection and AI-aware SOC operations

Organizations without AI and automation in security operations paid ₹316 million per breach on average. Gruve’s AI-aware SOC operations apply AI and automation to alert triage, detection, and response. The SOC also monitors AI systems and agents, not only traditional infrastructure. Furthermore, this continuous monitoring supports CERT-In’s emphasis on continuous exposure management.

Data security posture management and compliance mapping

Data security posture management discovers and classifies sensitive data before AI systems consume it. Gruve maps findings to DPDP, RBI FREE-AI, SEBI, and CERT-In expectations. This mapping addresses the safeguards duty behind the ₹250 crore DPDP penalty ceiling.

AI red teaming and continuous validation

Offensive security practices cut breach costs by ₹24.7 million on average in India. Gruve’s red teams test models for prompt injection, data leakage, and excessive agent permissions. Findings arrive with prioritized fixes and retest evidence for the next board meeting.

Evidence and reporting for boards and regulators

Boards and regulators need documented proof, not verbal assurances from technical teams. Gruve provides audit-ready logs, control mappings, and quarterly dashboards aligned to board metrics. These outputs also support the 72-hour DPDP reporting window and any regulator inquiry.

Boards that want a practical starting point can request an AI security readiness assessment from Gruve. With the partner role clear, the next section summarizes the actions each leader should take this quarter.

Key takeaways for CISOs and business leaders

The actions below are small enough to start this quarter and large enough to matter.

What boards should do this quarter

• Name one committee as the accountable owner for AI and cyber risk oversight.
• Request a complete AI inventory and a first quarterly AI risk dashboard.
• Schedule a tabletop exercise that tests the 72-hour DPDP reporting window under pressure.

What CISOs should do this quarter

• Publish the AI risk register with named owners, risk tiers, and review dates.
• Launch shadow AI discovery and offer approved tools as safe alternatives for employees.
• Translate the top three AI risks into estimated rupee losses for the board.

What business leaders should do this quarter

• Fund AI security as a growth enabler, using breach cost data to justify it.
• Name a business owner for every AI use case in your unit.
• Review vendor contracts for audit rights, AIBOM disclosure, and incident notification terms.

AI will keep moving faster than any regulation that tries to govern it. The boards that succeed will ask sharper questions this quarter and demand evidence by the next. Trust, in the end, belongs to the organizations that can prove their AI is under control.