Frequently asked questions
RBI, DPDP, EU AI Act, deployment, security, and how Zytra works. Don't see your question? Ask us.
What does SEBI's responsible AI consultation require for intermediaries?
SEBI's June 2025 consultation on responsible AI sets six principles for all SEBI-registered intermediaries using AI/ML: Fairness (no discriminatory outcomes), Accountability (named human accountability for AI decisions), Transparency (investors informed when AI influences decisions about them), Robustness (reliable performance across market conditions), Privacy (DPDP Act compliance at the AI inference layer), and Explainability (material AI-driven decisions explainable to investors). The principles apply to all entity types — stockbrokers, investment advisers, research analysts, portfolio managers, exchanges, and depositories. Full SEBI responsible AI control mapping →
What are RBI's eMRM guidelines and what do they require?
RBI's draft enhanced Model Risk Management (eMRM) guidelines, published in June 2026, establish a comprehensive validation lifecycle for all AI/ML models at regulated entities — banks, NBFCs, and financial companies. Key requirements: independent validation (even for third-party/vendor AI models), a four-tier model classification system (Critical, Significant, Standard, Minimal), ongoing performance monitoring with defined thresholds, and formal model retirement processes. Unlike the 2024 AI Circular (governance architecture) and FREE-AI Framework (quality standards), eMRM governs the validation lifecycle — how every model is assessed and managed throughout its production life. Full eMRM requirements guide →
What is the RBI FREE-AI Framework?
RBI's FREE-AI Framework (August 2025) sets 26 recommendations for Fairness, Reliability, Explainability, and Ethics in AI for all regulated entities. It is the implementation layer for responsible AI — specifying what banks must actually do, not just what they must have in policy. No AI governance platform other than Zytra has published a clause-level mapping of all 26 recommendations to controls. Full FREE-AI mapping →
Does RBI's 2024 AI circular apply to my bank?
Yes. RBI's 2024 guidance on Artificial Intelligence applies to all regulated entities — banks, NBFCs, housing financiers, and payment companies. It mandates board-approved AI governance frameworks, audit trails, AI inventories, outsourcing controls for third-party AI, and incident reporting for every AI system in production. Non-compliance creates examination exposure, not just a policy gap. Full compliance checklist → · How Zytra maps to each RBI obligation →
What does DPDP Act 2023 require for AI systems?
India's DPDP Act applies at the AI inference layer, not just data storage. Key obligations: consent for AI-driven profiling, purpose limitation for personal data used in training or inference, PII detection and redaction in model outputs, and data localisation for AI workloads. Banks using AI for credit scoring, KYC, or customer service must map these to their inference pipelines. Zytra is the only AI governance platform to have mapped DPDP Act obligations at the inference-pipeline level. Full DPDP coverage →
What does EU AI Act high-risk classification mean for banks?
AI used for credit scoring, insurance risk, fraud detection, or employment decisions is high-risk under the EU AI Act. This triggers Articles 9–15: risk management systems, data governance, technical documentation, logging, human oversight, accuracy and robustness. Non-compliance: up to €35M or 7% of global turnover. Zytra's EU AI Act pack covers Art. 9–15 and GPAI requirements (Art. 53–55), with automated evidence collection and pre-formatted conformity assessment documentation.
Can Zytra run in our own tenant?
Yes. Zytra deploys inside your own cloud account (AWS, Azure or GCP) or VPC, on-premise, or even air-gapped. Your data, prompts and model traffic stay within your perimeter, Zytra never sees your data. A fully managed cloud option is also available when you want the fastest setup.
Where does our data live?
Wherever you choose. In-tenant and on-prem deployments give you full data residency in the region and jurisdiction you require, important for RBI, DPDP, GDPR and internal bank policy.
Do you train on our data?
No. We practice data minimization and do not train models on your prompts, documents or outputs.
What security certifications do you have?
Zytra's controls are aligned to SOC 2 Type II, ISO 27001 and ISO 42001, with encryption in transit and at rest, SSO/SAML, RBAC and immutable audit trails. Our security package, report, DPA, pen-test summary, architecture and sub-processors, is available on request under NDA. See the Trust Center.
Which regulations do you support?
Pre-built compliance packs cover the EU AI Act, NIST AI RMF, ISO 42001 and SOC 2, plus India's RBI, SEBI and DPDP requirements. Map controls once and satisfy multiple frameworks.
Which models and LLMs does Aval work with?
Aval is model-agnostic, it sits as a guardrail layer in front of any LLM (commercial or open) with zero changes to your models. It governs at the model, agent, application and network layers.
How fast is Aval?
11.6ms inference on a single RTX 4090 at 184M parameters, with a 0.5% false-positive rate, fast and precise enough for real-time customer and agent workflows.
How long does it take to deploy?
Managed cloud can be operational in minutes with pre-built accelerators and the SDK. In-tenant and on-prem deployments depend on your environment and security review; our team supports the full process.
How does pricing work?
Plans scale from first deployment to enterprise, priced on scale and deployment model. See Pricing or talk to sales for a tailored quote.
How is Zytra different from Arthur AI, Credo AI, or Holistic AI?
Arthur AI, Credo AI, and Holistic AI cover EU AI Act, NIST AI RMF, and ISO 42001 — the three frameworks in every vendor deck. None have documented coverage of RBI's 2024 AI circular, SEBI's algorithmic AI oversight guidance, or DPDP Act 2023. Zytra maps all three at clause level with examination-ready evidence. Zytra also provides SR 11-7 MRM workflow engine, OSCAL export, signed evidence ledger, and 22 BFSI-specific guardrail categories — capabilities absent from all three competitors. Full comparison →
Who is behind Zytra?
Zytra is backed by Pennant Technologies Private Limited, whose pennApps Lending Factory is trusted by banks and NBFCs across Asia, so Zytra is built by a team that understands regulated banking from the inside.
Still have questions?
Our team is happy to walk through security, deployment and compliance in detail.