Aller au contenu principal
StrideRecruit

Security Engineer for Agentic AI

BRAHMA AI

Source : Himalayas — flux autorisé

Lieu
Royaume-Uni
Type de contrat
Temps plein · Télétravail
Expérience
Senior
Publiée
14 nov. 2026
Dernière fois vu actif
14 nov. 2026
Source
Himalayas
Employeur
BRAHMA AI
Voir sur Himalayas et postuler ↗

Ce lien ouvre l'offre sur Himalayas, où la candidature se poursuit. Il n'est pas confirmé qu'il s'agisse du site de l'employeur.

Description du poste

La traduction automatique n'est pas activée. La description est affichée dans sa langue d'origine.

Brahma AI operates at the intersection of enterprise Media Asset Management (MAM) and cutting-edge generative media. We build and scale industry-leading generative AI models, including hyper-realistic digital humans (ATMAN) and multilingual voice synthesis (VAANI), for world-class enterprise clients in entertainment, sports, healthcare, and retail.

Role Overview
We are looking for a Security Engineer for Agentic AI to make the agents, tools, and models Brahma teams rely on safe by default. You will own the security of our sovereign agent gateway for model routing, MCP tool hosting, and multi-agent orchestration: how agents authenticate, what they can reach, how they handle untrusted input, and how every action is logged.

This is an application and AI security role, not a cloud security role. Our Cloud Security Engineer owns GCP, Kubernetes, network, and infrastructure-as-code posture. You own the controls above it, in agents, tools, models, and their data flows, and you share secrets, sandbox runtimes, and security telemetry with them. You work as part of a team led by the Lead Engineer for Agentic AI, together with the AI Architect , Senior Engineers, ML Engineers and Product Management. You also work closely with Security and compliance and other teams using or contributing to Agentic AI. You own and build the security controls; the Architect designs security into the architecture, and Senior Engineers build to your standards and fix findings in their components.

This is an individual contributor role: building controls, tooling, and tests, and reviewing code and designs. You also share security knowledge within the team and with other Brahma teams, so safe practices spread beyond the people who build the controls. You follow emerging agent and AI threats, test new attack classes in bounded experiments, and bring proven defences into production. Once they are live, you own them: monitoring, detection, and response in an enterprise environment. We assemble our setup from open protocols, open-source components, and our own services, so you build what is missing and maintain what you run. There is no formal on-call rotation yet, but one may be introduced.

Key Responsibilities

1. Threat Modelling & Secure Design

  • Agentic Threat Modelling: Threat model agents, tools, MCP servers, and multi-agent workflows using the OWASP Top 10 for LLM Applications, OWASP Agentic AI threats, and MITRE ATLAS, and run a lightweight pre-launch review gate.
  • Secure-by-Default Reviews: Review agentic code and designs, publish secure defaults, and make risk trade-offs explicit and owned.
  • Security Knowledge Sharing: Teach secure agent practices within the team and to other Brahma teams through guides, threat-model walkthroughs, demos and office hours.

2. Identity, Isolation & Guardrails

  • Agent & Tool Identity: Design authentication and authorisation for agents and tools with OAuth 2.0/OIDC, token exchange, and scoped, short-lived credentials, with per-tool least privilege, fail-closed MCP authorisation, and secrets kept out of agent runtimes.
  • Sandboxed Execution: Harden the containers and sandboxes where coding agents and tool calls run, with filesystem isolation, egress control, and human approval for irreversible actions.
  • Guardrails & Policy Enforcement: Build gateway defences against direct and indirect prompt injection, tool abuse, excessive agency, and data exfiltration: input and output filtering, PII/DLP, and policy-as-code, tuned so they do not stall developers.

3. Adversarial Testing & Supply Chain

  • AI Red-Teaming: Build an automated adversarial test suite (injection, tool misuse, exfiltration, goal drift) that runs in CI and on a schedule, so we can show controls work, not only that they exist.
  • Supply-Chain Security: Govern which MCP servers, skills, plugins, models, and dependencies we adopt, through review, allow-listing, signing, SBOMs, and scanning.

4. Detection, Response & Compliance

  • Audit Logging & Detection:Make every model, tool, and agent call traceable to a person and an agent, and feed guardrail events and anomalies into security monitoring and alerting.
  • Production Security Ownership: Own the security controls once they are live: monitor their effectiveness, tune guardrails from production data, and keep them patched and current as models, tools and standards change.
  • Incident Response & Compliance: Own the playbook for agent-related incidents and lead the technical response. Map controls to ISO 27001, ISO/IEC 42001, SOC 2, GDPR, and TPN, and keep evidence audit-ready.

Must Haves

  • Security Engineering Background: 5+ years in security engineering with a strong application or product security component, and fluent Python.
  • Build, Not Only Audit: Track record of shipping security tooling, controls, and automated tests as production code.
  • LLM & Agent Security: Hands-on experience securing LLM applications or agents against prompt injection and unsafe tool use.
  • Identity & Authorisation:Deep knowledge of OAuth 2.0, OIDC, token exchange, workload identity, and least-privilege design.
  • Threat Modelling:Experience leading structured threat modelling and turning findings into prioritised engineering work.
  • Secure SDLC & DevSecOps: Experience embedding security scanning and checks into CI/CD, and securing containerised workloads.
  • Risk Communication: Ability to explain risk and trade-offs clearly to engineers and leadership, and to document decisions.
  • Knowledge Sharing:Track record of teaching and documenting security practices for colleagues and other teams.

Nice to Have

  • AI Red-Teaming Tools: Experience with garak, PyRIT, promptfoo, or similar.
  • Policy Engines: Practical use of OPA/Rego, Cedar, or similar.
  • Sandboxing Technology: Familiarity with gVisor, Firecracker, seccomp, or Landlock.
  • Agent Infrastructure:Experience building or securing MCP servers, LLM gateways, or guardrail services.
  • GCP:Working knowledge of IAM, Workload Identity, Secret Manager, and Cloud Logging.
  • AI Governance & Media Security: Exposure to ISO/IEC 42001, NIST AI RMF, or MPA/TPN content security.
  • Certifications: OSCP, GIAC, CSSLP, or equivalent.

Originally posted on Himalayas

Offres fournies par Himalayas · himalayas.app

À propos de cet emploi

Cette page décrit une offre en cours : Security Engineer for Agentic AI chez BRAHMA AI, indiquée comme Royaume-Uni.

Frequently asked questions

Où est basé cet emploi ?
Cet emploi est indiqué comme Royaume-Uni.
De quel type de travail s’agit-il ?
Il s’agit d’un poste Temps plein · Télétravail chez BRAHMA AI.
Quand cet emploi a-t-il été publié ?
Cet emploi a été publié le 14 nov. 2026.
Où postuler à cet emploi ?
Utilisez le lien de candidature sur cette page. Il ouvre l’offre originale sur Himalayas, où vous postulez — StrideRecruit ne reçoit pas les candidatures.
Signaler cette annonce

Dites-nous si cette annonce semble incorrecte. Nous examinons les signalements ; un signalement ne supprime pas l’annonce à lui seul.

Nous ne demandons ni votre nom ni votre e-mail et ne conservons pas votre adresse IP avec le signalement.