21 Nov
|
Flexiple
|
Delhi
Hiring for Lean SuperIntelligence — via Flexiple, the recruiting partner on this role.
About the Company
LSI (Lean SuperIntelligence) is a security AI lab building self-improving cybersecurity LLMs. LSI-Offense-1 finds, exploits and validates vulnerabilities; LSI-Defense-1 detects, investigates and responds to attacks. The two train against each other inside LSI Cyber World Models, realistic instrumented enterprise environments where every outcome is verified, and they run on-prem or air-gapped inside the enterprise that owns them.
LSI is founded by the team behind India's first sovereign LLM and AI labs at Alexa, Uber AI and Salesforce Research, with elite security researchers, and is backed by Accel, Fusion Fund, SVQ, Inventus Capital and Dreamit Ventures, plus angels who led AI and security at OpenAI, CrowdStrike, SentinelOne and Zscaler.
About the Role
You will architect LSI's agentic security system: a team of AI agents that attack, defend and harden a customer's estate continuously. Offense agents probe like an attacker, defense agents detect, investigate and respond like the best responder, and remediation agents turn findings into fixes. You own how they reason, act, hand off to each other and stay under human control.
You are fluent on both sides: you have run adversary simulations and led investigations, and you have built agents that do both. You will turn that into system architecture, training signal and evaluations, working with ML researchers, offense and defense engineers and design partners.
What you'll do
- Architect the multi-agent system: offense, defense and remediation agents, the orchestrator that routes work between them, and the shared context that carries a finding from attack path to detection to fix.
- Design the harness and actuators agents use across endpoint, identity, cloud, SaaS, code and CI/CD, and SIEM/XDR/SOAR, with every action scoped,
traceable and replayable.
- Build the security context graph that unifies endpoint, identity, cloud, app and agent signals with threat intel, past incidents and policy.
- Own routing between LSI's specialised cyber models and frontier models, trading accuracy, latency and cost per task.
- Define agent identity, least privilege, autonomy levels, human sign-off for high-impact actions, audit trails and rollback.
- Own one evaluation scorecard across find, exploit, detect, investigate, triage and fix, and lead staff engineers and design-partner pilots.
What you bring
- 10+ years with hands-on depth in both offense (red teaming, adversary simulation, pentest, AppSec, vulnerability research) and defense (detection engineering, threat hunting, incident response, threat research).
- Has designed and shipped a security platform, detection and response system or SOC used at enterprise or cloud scale.
- Has built production LLM agents or multi-agent systems for security with tool use, orchestration and measured results, and knows how to catch hallucinated findings.
- Has led full-kill-chain adversary simulations and investigations into sophisticated intrusions.
- Depth in SIEM/XDR/SOAR, AWS/Azure/GCP, Entra/Okta, Kubernetes, large-scale log pipelines and security graphs.
- Has designed human-in-the-loop controls, service identity and permissions, or guardrails for automated response.
- Fluent in Python plus one of Go, Rust or TypeScript.
Nice to have
- RL environments, reward functions, verifiers or post-training for agentic tasks.
- Built agentic SOC, autonomous pentest or exposure-management products at a security vendor or AI lab.
- Security of AI agents: prompt injection, tool and MCP abuse, agent identity (OWASP LLM / Agentic Top 10, MITRE ATLAS).
- Public CVEs, open-source detection content, or talks and papers at Black Hat, DEF CON, USENIX Security or IEEE S&P.;
Originally posted on: 20 Nov 2026
📌 Principal Architect, Frontier AI Offense & Defense Security (Delhi)
🏢 Flexiple
📍 Delhi