Lean SuperIntelligence Debuts Self-Improving Cybersecurity Models

The company’s LSI-Offense-1 and LSI-Defense-1 models operate in a continuous loop: the offensive unit identifies and validates vulnerabilities, while the defensive unit learns from those findings to refine its detection and response capabilities. According to the firm, this architecture allows the systems to resolve security threats in seconds, significantly faster than traditional manual investigation methods. Independent benchmarks indicate that these models surpass industry-standard systems like Claude Mythos, GPT-6, and Gemini-3.8-Cyber in both vulnerability discovery and multi-step threat detection.

Founder and CEO Chandra Khatri, whose background includes leadership roles at Krutrim and Got It AI, emphasizes that the platform solves the enterprise dilemma of renting intelligence from cloud providers. By deploying air-gapped, on-premise models, organizations retain control over their proprietary data while benefiting from a security posture that evolves alongside the threat landscape. The startup has secured pre-seed funding from investors with expertise in companies such as OpenAI, CrowdStrike, and SentinelOne. LSI is currently granting access to a select group of design partners, including managed security service providers and Fortune 500 enterprises, to facilitate sovereign deployment.