AI-as-a-Service Providers Vulnerable to PrivEsc and Cross-Tenant Attacks: Cyber Risks and Legal Recourse in India

Introduction: The Double-Edged Sword of AI-as-a-Service (AIaaS)

The rapid commercial adoption of generative artificial intelligence and large language models (LLMs) has led to an explosion of AI-as-a-Service (AIaaS) platforms. Instead of building massive compute clusters from scratch, enterprises in India—from fintech startups in the Bandra-Kurla Complex (BKC) to manufacturing powerhouses in Thane—rely on third-party cloud AI vendors for model hosting, fine-tuning, Retrieval-Augmented Generation (RAG), and real-time inference APIs.

However, this shared-infrastructure model introduces grave architectural vulnerabilities. Recent cybersecurity research has highlighted that multi-tenant AI infrastructure is increasingly susceptible to Privilege Escalation (PrivEsc) and Cross-Tenant Attacks.

When these attacks succeed, an attacker sharing the same cloud or inference cluster can break isolation barriers, elevate their privileges, inspect another enterprise's proprietary prompts, exfiltrate private training data, or manipulate vector embeddings. For Indian companies handling sensitive commercial IP or personally identifiable information (PII), a cross-tenant compromise is not merely an engineering incident—it is a full-blown regulatory, civil, and criminal crisis.

Understanding the Threat: PrivEsc and Cross-Tenant Exploits in AI

To address the legal and technical aftermath, legal teams and CISOs must first understand how these exploits materialize.

1. Privilege Escalation (PrivEsc) in AI Workflows

Privilege escalation occurs when an adversary exploits a bug, design flaw, or configuration error to gain elevated permissions within an application or cloud infrastructure beyond what was designated.

In an AIaaS environment, this often manifests via:

Agentic Tool Misuse & Indirect Prompt Injection: An attacker crafts adversarial inputs that compel an AI agent to execute unauthorized internal APIs or cloud functions with service-account privileges.
Container & Kernel Breakouts: Many AIaaS platforms execute custom code or Python sandboxes for data cleaning and model evaluation. Vulnerabilities in container runtimes or outdated Linux kernels allow attackers to escape the isolated container and gain root access to the host node managing GPU clusters.
Misconfigured IAM Roles: Over-permissive Identity and Access Management (IAM) configurations in cloud setups (e.g., AWS, Azure, GCP) allow lower-tier user roles to assume administrative or model-orchestration credentials.

2. Cross-Tenant Infiltration

Most AIaaS providers operate on a multi-tenant model to optimize expensive GPU infrastructure. Cross-tenant attacks exploit the failure of multi-tenancy isolation:

Vector Database Contamination: In multi-tenant RAG architectures, improper namespace isolation within vector databases can allow Tenant A to query or extract vector embeddings and contextual documentation belonging to Tenant B.
Shared Model Cache Poisoning: Shared model caches or inference endpoints can leak past inputs, outputs, or fine-tuning weights between independent enterprise accounts.
Side-Channel GPU Attacks: Malicious tenants sharing physical GPUs can sometimes monitor memory access patterns and power states to infer proprietary model weights or tokens processed by other users on the same physical hardware.

Indian Legal Framework Governing AIaaS Data Breaches

When an AIaaS platform's multi-tenancy fails and Indian businesses suffer unauthorized data access or disruption, multiple statutory frameworks are triggered.

1. Information Technology Act, 2000 (IT Act)

Section 43: Deals with unauthorized access, downloading, extracting, or damaging computer systems or networks. Victims can seek civil compensation for damages before the Adjudicating Officer appointed under the IT Act.
Section 66: Criminalizes hacking and computer-related offences defined under Section 43 with imprisonment up to three years or fines up to five lakh rupees.
Section 66C & 66E: Applicable when access credentials, tokens, or private customer records are illegally accessed or published.
Section 72A: Imposes criminal liability for disclosure of information in breach of lawful contract.

2. Bharatiya Nyaya Sanhita, 2023 (BNS)

Following the overhaul of criminal jurisprudence in India, relevant penal provisions for cyber sabotage, identity theft, and corporate fraud fall under the Bharatiya Nyaya Sanhita (BNS):

Section 303 (Theft): Prosecuting the intentional extraction and dishonest misappropriation of proprietary digital assets, code, or business data.
Section 316 & 318 (Cheating): Deceptive practices used by rogue actors or fraudulent service providers falsely asserting security compliances.
Section 61 (Criminal Conspiracy): Applicable where internal actors collude with external attackers to bypass isolation controls.

3. Digital Personal Data Protection Act, 2023 (DPDPA)

If the cross-tenant breach exposes personal data of Indian citizens:

Obligations of Data Fiduciaries: Indian companies using AIaaS remain the "Data Fiduciaries." Even if the breach happened within the third-party AI platform (the "Data Processor"), the primary liability to implement reasonable security safeguards rests with the Fiduciary.
Mandatory Breach Notification: Under the DPDPA, any personal data breach must be reported to the Data Protection Board of India (DPBI) and the affected individuals.
Severe Financial Penalties: Failure to take reasonable security safeguards to prevent personal data breaches can attract penalties up to ₹250 crore.

4. CERT-In Directions (Cybersecurity Directions 2022)

Under Section 70B of the IT Act, the Indian Computer Emergency Response Team (CERT-In) mandates that cyber incidents must be reported within 6 hours of noticing or being informed of them. Incidents involving unauthorized access to systems, data breaches, and compromise of critical cloud infrastructure mandatorily fall within this window.

Evidentiary Preservation Under the Bharatiya Sakshya Adhiniyam, 2023 (BSA)

In the event of an AIaaS security compromise, Indian enterprises must preserve digital evidence according to the legal standards of the Bharatiya Sakshya Adhiniyam, 2023 (BSA). Section 63 of the BSA replaces the erstwhile Section 65B of the Indian Evidence Act for the admissibility of electronic records.

To ensure audit logs, API telemetry, and vector store queries stand up in Indian courts:

Secure Immutable Logs: Immediately freeze and snapshot all cloud access logs, container audit trails, and API access tokens.
Hash & Timestamp: Create cryptographic hashes (SHA-256) of captured disk snapshots and log files.
Prepare the Electronic Record Certificate: Ensure system administrators or cloud security officers are ready to execute the mandatory certificate under Section 63 of the BSA verifying the lawful custody and operational integrity of the computer system during log creation.

Actionable Response: How Victims in the Mumbai Metropolitan Region (MMR) Can Report

For businesses operating in Mumbai, Thane, or Mira-Bhayandar that identify an active cross-tenant data leak or unauthorized privilege escalation:

Step 1: Immediate Technical and Internal Mitigation

Revoke compromised API keys, rotate federated IAM tokens, and terminate suspicious container sessions.
Notify your third-party AI provider formally to preserve host-level hypervisor and orchestrator logs.

Step 2: Regulatory Notification

File an incident report with CERT-In via their reporting incident form or incident email (incident@cert-in.org.in) within the 6-hour timeline.

Step 3: Filing Cyber Complaints in Mumbai / MMR

Victims should register complaints through national and local jurisdictional cyber crime cells:

National Portal: Log an initial complaint at cybercrime.gov.in or call the Cyber Helpline 1930.
Jurisdictional Cyber Police Stations in MMR:
- Mumbai (South / Central / West): Visit the Cyber Police Station at the Bandra-Kurla Complex (BKC) or the regional cyber cells (e.g., Worli Cyber Cell for South Mumbai).
- Thane Jurisdiction: Approach the Thane Cyber Crime Police Station located near the Commissioner of Police office, Court Naka, Thane (West).
- Mira Road & Bhayandar (MBVV Commissionerate): Approach the Mira-Bhayandar Vasai-Virar (MBVV) Cyber Crime Police Station to lodge a formal written representation and obtain a First Information Report (FIR).

Frequently Asked Questions (FAQs)

1. If our AIaaS vendor suffers a cross-tenant leak, are we liable under the DPDPA 2023?

Yes. Under the DPDPA 2023, the entity determining the purpose and means of processing personal data is the Data Fiduciary. The AI vendor is typically a Data Processor. The law holds the Data Fiduciary accountable for failing to prevent a personal data breach, regardless of whether the processing was outsourced. A robust contract and verifiable security audits of your vendor are critical to mitigating legal exposure.

2. Can an enterprise file a civil claim against a foreign AIaaS vendor in an Indian court?

It depends on the dispute resolution and jurisdiction clause in the Master Services Agreement (MSA). While foreign vendors often specify international arbitration (e.g., in Singapore or the US), Indian courts may exercise jurisdiction under the IT Act if harm is suffered on computer networks located in India. Having an Indian legal team review limitation-of-liability and choice-of-law clauses before onboarding AI platforms is crucial.

3. What is the immediate legal timeline for reporting an AI data breach?

Under CERT-In guidelines, cyber incidents involving unauthorized system access or data breaches must be reported within 6 hours of detection. Under the DPDPA, notifications to the DPBI and affected data principals must be made promptly as prescribed by future Board rules.

4. What penal charges apply if a former employee uses PrivEsc to siphon AI data?

Such actions are punishable under Section 43 read with Section 66 and Section 66C of the Information Technology Act (hacking and identity theft), as well as Section 303 (Theft) and Section 316 (Cheating/Breach of Trust) of the Bharatiya Nyaya Sanhita, 2023.

Legal Advisory and Cyber Incident Support from APS Law

Navigating emerging cyber risks in artificial intelligence requires a proactive blend of technical comprehension and aggressive statutory compliance. At APS Law, our cyber law and commercial litigation practice assists startups, corporate enterprises, and technology service providers across Mumbai, Thane, and Mira Road.

Whether you need to review AIaaS vendor contracts, navigate CERT-In compliance, prepare for DPDPA adherence, or file urgent cyber complaints before law enforcement authorities, our team provides focused, practical legal counsel.

Facing a cybersecurity vulnerability or data breach? Contact APS Law today to request a legal consultation.