AI Security Architect - Applied AI Security

As we deploy AI agents across trading, risk, compliance, and customer systems, we’re building the security architecture that keeps those systems trustworthy at scale.

This isn’t a compliance checkbox role. You’ll architect the security layer for systems where a compromised AI agent can authorize transactions, move funds, or access millions of user records. The threat model includes nation-state actors, sophisticated financial fraud, and the exact class of autonomous AI attack campaigns that are actively targeting financial infrastructure today.

You’ll work directly with Group Risk & Security leadership to define how we build, deploy, and operate AI safely - from model selection to runtime monitoring to red team automation., AI System Security Architecture

• Design end-to-end security architecture for AI stack: LLM inference pipelines, AI agents, MCP-based tool orchestration, and agentic workflows
• Define trust boundaries, authorization models, and least-privilege principles for multi-agent systems
• Architect the AI Security Gateway: prompt/response inspection, rate limiting, anomaly detection, circuit breaking across all AI-powered products

Threat Modeling & Red Teaming

• Build and operate an AI-specific threat model covering prompt injection, jailbreak, model inversion, supply chain poisoning, and autonomous agent misuse
• Lead red team exercises against our AI deployments; evolve our Red vs Blue automation platform (currently in production, LLM-driven)
• Define detection signatures for AI-assisted attacks - we’ve seen the Hermes/Strix/Cairn class of autonomous attack campaigns firsthand

Secure AI Development Lifecycle (AI-SDL)

• Define security review processes for AI features: model selection, fine-tuning pipelines, RAG system design, tool call authorization
• Evaluate third-party AI vendors and integrations (model APIs, vector DBs, agent frameworks) for security posture
• Work with product and infra teams to embed security requirements from design through deployment

TEE & Confidential Computing

• Architect TEE-based deployments for sensitive AI inference (AMD SEV-SNP, Nvidia Confidential Computing)
• Design attestation flows, measurement coverage, and post-deployment verification - with the rigor of a Trail of Bits-level audit, * The threat surface is real. AI-powered autonomous attack campaigns are already hitting financial infrastructure. You won’t be building defenses for hypothetical future threats - the Cairn-class agent that chains SQLi * RCE * AWS secrets extraction in one autonomous run is already operational against our industry.
• The scope is unusually large. We run AI across risk scoring, fraud detection, customer support, compliance screening, and internal tooling. The security architecture you design will cover all of it.
• You’ll ship, not just advise. This role produces systems, not slide decks. Red team automation platform, AI security gateway, TEE attestation flows - built and running in production.
• Becoming a product. Our security capabilities are on a path to commercialization. The AI security architecture you build here will eventually be offered as a product to other exchanges and financial institutions.

• 6+ years in security engineering or architecture; 2+ years focused on ML/AI systems
• Deep understanding of LLM internals: pretraining, RLHF, inference, safety alignment mechanisms - and how each layer can be attacked
• Hands-on experience with prompt injection, jailbreak research, or adversarial ML (published work, CTF, or production red team)
• Experience building or evaluating security for production AI systems (not just reading papers)
• Familiarity with AI agent frameworks: LangChain, CrewAI, OpenAI Assistants, Anthropic’s tool use, MCP

Strong Signal

• Published red team research, CVEs in AI tooling, or presented at Black Hat / DEF CON / IEEE S&P on AI security topics
• Experience with TEE deployment in production (SGX/TDX/SEV-SNP)
• Background in financial services, crypto, or other high-stakes transaction systems
• Contributed to AI safety frameworks (responsible disclosure, model cards, safety evals)