Senior AI Security Engineer

Position: Senior AI Security Engineer

Experience: 5+ years

Location : Pune

Key Responsibilities:

The "Red" (Adversarial Simulation)

AI Red Teaming: Conduct advanced adversarial testing on Large Language Models (LLMs) and Agentic AI workflows. Execute prompt injections, jailbreaking, model inversion, and data poisoning attacks.

Agentic Threat Simulation: Test autonomous agents for "excessive agency" vulnerabilitiesmanipulating agents into performing unauthorized actions (e.g., executing SQL commands, escalating privileges, or leaking PII).

Automated & Manual Testing: Leverage tools like Garak, PyRIT, or TextAttack for automated scanning, while applying manual creativity to find logic flaws in multi-agent orchestration.

Chain-of-Thought Exploitation: Analyze and exploit flaws in the reasoning loops of autonomous agents (e.g., LangChain or AutoGen workflows).

The "Blue" (Defense & Engineering)

Guardrail Engineering: Design and implement input/output filters using tools like NVIDIA NeMo Guardrails, Llama Guard, or Lakera.

Identity & Access Control: Architect "Non-Human Identity" policies for AI agents, ensuring they adhere to Least Privilege (e.g., preventing an agent from deleting DB records).

Detection Engineering: Build monitoring pipelines to detect real-time attacks (e.g., identifying a "DAN" attack pattern in live chat logs) and automate response triggers.

Remediation: Don't just report bugsfix them. Rewrite system prompts to be robust against social engineering and re-architect RAG pipelines to prevent data leakage.

The FDE (Client Engagement)

Embedded Problem Solving: Work on-site with client engineering teams to understand their specific business logic and deploy secure AI architectures.

Threat Modeling: Lead workshops to map the "Blast Radius" of a client's AI agents (i.e., if this agent is compromised, what can it destroy?).

Requirements

Skills and Qualifications:

Experience: 5+ years in Cybersecurity, with at least 2 years focused on Application Security, Penetration Testing, or ML Security.

AI/ML Depth: Deep understanding of LLM architectures (Transformers, RAG, Fine-tuning). You understand how a model "thinks" and where it hallucinates.

Technical Stack:

Languages: Proficient in Python (mandatory for building custom attack scripts and harness).

AI Frameworks: Experience with LangChain, Semantic Kernel, or Bedrock.

Security Tools: Burp Suite, OWASP ZAP, plus AI-specific tools (Garak, PyRIT). Offensive Mindset: Proven ability to think like an adversary (e.g., CVEs, Bug Bounties, or CTF wins).

Defensive Engineering: Experience implementing WAFs, API Gateways, or IAM policies (OAuth, OIDC)

Nice to Have:

Experience with Agentic Identity concepts (SPIFFE/SPIRE, Machine ID).

Certifications: OSEP, OSWE, or specific AI Security certifications (e.g., NVIDIA, SANS).

Contribution to open-source AI security projects or OWASP Top 10 for LLM.