Purpose-built attacker intelligence and community-extensible attack library empower CISOs, VPs of AI, and CIOs to continuously red team autonomous AI systems at enterprise scale delivering production-grade resilience before threats escalate.

SAN FRANCISCO, March 19, 2026 (GLOBE NEWSWIRE) -- Votal AI, the AI-native security platform purpose-built for agentic AI systems and founded by cybersecurity veterans Bobby Gupta (CEO) and Jyotirmoy Sundi (CTO), today announced two groundbreaking capabilities in its Continuous Agentic Red Teaming (CART) platform: an RLHF-trained adversarial attacker model and the open-sourcing of its comprehensive Attack Catalog.

These launches come just days before RSA Conference 2026 (March 23?26, Moscone Center, San Francisco) the world's largest and most influential cybersecurity event, where Votal AI will showcase live demonstrations of CART simulating multi-stage adversarial campaigns against production agentic AI systems. The timing enables security leaders to evaluate and adopt continuous red teaming solutions amid surging agentic AI adoption and escalating risks.

Agentic AI systems now autonomously orchestrate tools, query data, execute transactions, and make decisions across enterprise environments expanding the attack surface far beyond traditional LLMs. A single successful jailbreak or tool misuse can lead to unauthorized API calls, data exfiltration, cross-tenant contamination, or compliance violations. Legacy point-in-time red teaming falls short against non-deterministic, autonomous agents.

Votal AI's CART platform closes this gap with automated, continuous adversarial testing tailored for the agentic era delivering actionable intelligence, compliance mapping, and remediation at scale.

1. RLHF-Trained Adversarial Attacker Model Fine-tuned via reinforcement learning from human red team experts, this model learns from real bypass outcomes not just prompts to generate adaptive, effective attacks. It navigates CART's seven-stage Agentic AI Kill Chain (prompt injection → privilege escalation → reconnaissance → persistence via RAG/memory poisoning → C2 via tool misuse → lateral movement → actions on objective). Continuously retrained on emerging threats (research, CVEs, intelligence feeds), it evolves in lockstep with adversaries, providing CISOs and CIOs with evidence-based assurance for high-stakes deployments.

2. Open-Source Attack Catalog Votal AI is open-sourcing its structured Attack Catalog covering 35+ security categories, 185+ named techniques, 18 encoding/obfuscation types, and 8 multi-agent scenarios aligned with Pangea, CrowdStrike, OWASP LLM Top 10, NIST AI RMF, MITRE ATLAS, EU AI Act, GDPR, HIPAA, PCI-DSS, and more. Security teams, researchers, and developers can inspect, customize, and contribute vectors (reviewed for inclusion in CART), enabling vertical-specific extensibility (e.g., PHI leakage in healthcare, unauthorized transactions in finance, ICS manipulation in manufacturing).

Joint Statement from Founders "As agentic AI becomes critical infrastructure, the security imperative is clear: static or periodic testing is no longer sufficient. These systems make autonomous decisions with real-world consequences demanding continuous, adaptive red teaming. By releasing our RLHF-trained attacker model and open-sourcing the Attack Catalog, we're equipping CISOs, VPs of AI, and CIOs with transparent, community-powered tools to build resilient, compliant AI ecosystems from day one." - Bobby Gupta, CEO & Co-Founder and Jyotirmoy Sundi, CTO & Co-Founder, Votal AI

See CART in Action at RSA Conference 2026 Visit Votal AI at RSA 2026 for live demos of multi-stage attacks and defenses. Schedule a briefing: https://votal.ai/rsa-conference-2026/

Key Platform Highlights

• 100K+ dynamic attack prompts across 35+ categories
• RLHF-trained adaptive attacker with 20x faster throughput than human red teamers
• 30ms guardrail latency (industry-leading)
• 22 preset industry scan suites with 6 authentication types
• Full compliance mapping and audit-ready reporting
  
   

About Votal AI Votal AI delivers Continuous Agentic Red Teaming (CART) for LLM applications and autonomous AI systems combining RLHF-trained attackers, extensible catalogs, kill-chain sequencing, and closed-loop remediation. Headquartered in San Francisco, Votal AI serves regulated enterprises in healthcare, finance, manufacturing, and government. Learn more at votal.ai.

Media Contact: Aayush@votal.ai

Copyright 2026 GlobeNewswire, Inc.