What's new in Zytra
Product updates, new capabilities, regulatory coverage additions, and research releases.
Major milestones only. Minor patches and dependency updates not listed.
Agent Control Room Preview
Illustrative AI governance control room mockup showing real-time policy enforcement, D-series agentic threat monitoring (D-09 Tool Description Poisoning, D-10 A2A Trust Exploitation, D-11 Corrigibility & Goal Binding), and evidence generation across 6 sample AI assets. Preview available without connecting a live environment.
Preview →Regulatory Crosswalk v1.0
40 regulatory obligations across 7 frameworks: EU AI Act (High-risk + GPAI), SR 26-2 with Footnote 3 GenAI exclusion, RBI draft MRM Guidance (Jun 2026), NIST AI RMF, DPDP Act 2023, SEBI, and ISO 42001. Filter by framework, compliance status, and Zytra module. Every obligation links to its authoritative source.
View crosswalk →AGMM v1.0 — Agentic Governance Maturity Model
Published methodology for assessing agentic AI governance maturity across 5 domains and 4 levels. Covers planning oversight, tool invocation control, memory governance, multi-agent identity, and continuous monitoring. Freely available.
Read the methodology →RBI Draft MRM Guidance Coverage
Platform updated to cover RBI's draft Guidance on Model Risk Management (published 24 Jun 2026). Risk atlas obligations, control mappings, and evidence pack updated to reflect the draft guidance. Final RBI guidance coverage will follow on publication.
Aval v2.0 — 0.5% FPR at 11.6ms
Aval guardrails v2.0 achieves 0.5% false-positive rate across 22 BFSI threat categories at 11.6ms inference latency on RTX 4090. Trained exclusively on BFSI adversarial data. FinProof benchmark results published alongside the release.
View benchmark →KV Cache Governance + Forensic Replay
KV Cache Governance detects and prevents cross-tenant data leakage in shared LLM inference caches — the only platform to govern KV caches at the inference layer. Forensic Replay reconstructs the exact model context state at any point in time from cache access logs, enabling regulatory investigation and board-submittable evidence. Enterprise tier.
Agentic Security Module GA
Agentic security module generally available. 75 agentic risks across 11 domains, 44 dedicated controls, framework crosswalks to MITRE ATLAS and OWASP MAESTRO. Instruments at the orchestration layer — no agent-framework changes required. Compatible with LangGraph, LlamaIndex Workflows, CrewAI, and custom MCP stacks.
Agentic security →AIBOM — AI Bill of Materials
Automated AI Bill of Materials generation for every governed AI system. Captures model provenance, training data lineage, dependency chain, and deployment configuration. Exportable alongside OSCAL evidence packs. Enterprise tier.
OSCAL Export + Governance SDK v1.0
OSCAL-formatted evidence export for GRC platform integration (ServiceNow, Archer). Governance SDK adds policy gates to CI/CD pipelines — fail the build if a model doesn't clear its governance checkpoint before merge. Evidence is hash-chained and timestamped at export.
FinProof Benchmark Published
FinProof: the first BFSI-specific AI safety benchmark covering 14 adversarial categories (B-01–B-14). Benchmark openly published with methodology, evaluation protocol, and leaderboard. Any vendor can submit for independent evaluation.
Read FinProof →India Regulatory Pack v1.0
RBI, SEBI, and DPDP Act 2023 controls mapped to platform modules. India's AI governance framework spans three regulators with overlapping scope. The India pack unifies all three into one governance workflow with shared evidence collection.
Aval v1.0 GA — 22 BFSI Threat Categories
Aval guardrails v1.0 generally available. 22 BFSI threat categories: prompt injection, jailbreak attempts, PII exfiltration, regulatory breach, financial advice, legal advice, medical advice, brand risk, confidential data, malware, social engineering, over-refusal, policy violation, bias/discrimination, misinformation, copyright, credential harvest, CSAM, violence/threats, hate speech, competitor mention, and benign (negative class). BFSI-native training data.
Risk Atlas v1.0
Living AI risk atlas with 141+ risks across 11 domains, publicly browsable. Framework crosswalks to NIST AI RMF, OWASP LLM Top 10, MITRE ATLAS, and ISO 42001. Filter by severity, framework, capability pattern, and deployment type.
Explore →Platform GA — Governance, Guardrails, Red-team
Zytra AI governance platform generally available. Three pillars: Governance (risk atlas, control effectiveness scoring, residual risk heat maps, evidence ledger), Aval guardrails (runtime safety and threat detection), and Red-team (adversarial evaluation against FinProof benchmark).
Stay current on AI governance
Follow new regulatory coverage and platform capabilities.