Vacancy description
Metaforms
Bangalore, भारत
Metaforms builds AI infrastructure for research delivery teams. Our agents help market research agencies and sample providers execute complex workflows with greater speed, consistency, and scale.
As these agents interact with customer data, files, APIs, integrations, and other tools, security must be built into every layer of the platform. We are looking for an engineer who is excited about understanding how AI systems can fail—and building the controls that make them safe, reliable, and secure by default.
We’re looking for an AI Security & Platform Engineer to help secure the AI agents and infrastructure behind Metaforms.
You will test our AI systems adversarially, identify vulnerabilities across prompts, context, data, tools, integrations, and execution environments, and work with engineers to fix them properly. You will also build platform-level controls, automated evaluations, monitoring, and secure infrastructure that prevent similar vulnerabilities from recurring.
This is a hands-on engineering role, not a policy or compliance position. You will write code, inspect production systems, reproduce vulnerabilities, build security tooling, and contribute directly to product and infrastructure improvements.
You will work closely with engineering leadership and gradually take ownership of larger AI security and platform initiatives.
Adversarially test AI agents for prompt injection, context poisoning, unsafe tool use, data leakage, memory manipulation, and unintended behavior.
Identify security risks introduced through customer prompts, uploaded files, retrieved content, external links, integrations, and model-generated outputs.
Test whether agents can access unauthorized tools, records, tenants, credentials, or internal services.
Threat-model new AI features, agent workflows, model integrations, and tool-calling capabilities before they reach production.
Evaluate risks in RAG pipelines, context assembly, agent memory, system prompts, MCP servers, model providers, and third-party AI frameworks.
Build repeatable adversarial evaluations and regression tests for discovered vulnerabilities.
Design controls for least-privilege tool access, scoped credentials, sandboxing, runtime policy enforcement, and human approval.
Ensure model output is validated before it reaches databases, APIs, executable systems, or customer-visible workflows.
Improve auditability across model calls, retrieved context, tool invocations, agent decisions, and resulting actions.
Monitor emerging AI attack techniques and translate relevant risks into practical product controls.
Review customer-facing applications and APIs for authentication, authorization, tenant-isolation, injection, and business-logic vulnerabilities.
Secure customer-controlled payloads, file uploads, webhooks, exports, and third-party integrations.
Reproduce security reports, determine credible impact, and distinguish exploitable vulnerabilities from theoretical findings.
Work directly with engineers to implement production-ready fixes and regression tests.
Help build secure, reusable patterns for input handling, authorization, secrets, data access, and external integrations.
Perform focused code reviews, penetration tests, and architecture assessments for security-sensitive changes.
Build and improve the secure, reliable infrastructure behind model calls and agent workflows.
Harden cloud permissions, service identities, secrets, storage, networking, deployment pipelines, and production access.
Build guardrails into CI/CD and infrastructure-as-code so common security problems are caught before deployment.
Improve model-provider routing, retries, timeouts, fallbacks, quotas, rate limits, and cost controls.
Add observability for unusual model behavior, tool usage, data access, errors, latency, and token consumption.
Build containment mechanisms and kill switches for unsafe or misbehaving agents.
Support production debugging and the investigation of application, infrastructure, and AI-security incidents.
2–3 years of hands-on experience across AI engineering, backend/platform engineering, application security, infrastructure security, DevSecOps, or a related engineering role.
Experience building, operating, or securing production AI systems using LLM APIs, agents, tool calling, RAG, model gateways, or similar technologies.
Strong programming ability in Python, TypeScript, Go, or another relevant language.
Understanding of common web and API security risks, including authentication, authorization, injection, tenant isolation, secrets, and unsafe data handling.
Practical experience with cloud infrastructure, CI/CD, containers, monitoring, or infrastructure-as-code.
Ability to investigate a suspected vulnerability, reproduce it, assess its real-world impact, and contribute an effective fix.
Ability to reason about trust boundaries between users, models, customer data, application code, tools, and third-party services.
Comfort debugging systems through source code, logs, traces, metrics, and database queries.
Strong written communication and the ability to explain security risks clearly without unnecessary alarm.
A builder’s mindset: you enjoy implementing durable controls, not only identifying problems.
We care more about strong fundamentals, practical work, and learning velocity than matching every item in the description.
Hands-on experience with AI red teaming or adversarial model testing.
Familiarity with prompt injection, insecure output handling, tool-use vulnerabilities, agent sandboxing, or context leakage.
Experience securing multi-tenant B2B SaaS products.
Experience with OAuth, webhooks, file processing, MCP, and third-party integrations.
Experience building security test harnesses, fuzzers, automated evaluations, or internal security tools.
Familiarity with SAST, DAST, dependency scanning, secret scanning, or infrastructure scanning.
Experience with model gateways, multi-provider routing, AI observability, or LLM evaluations.
Participation in bug bounties, CTFs, security research, or relevant open-source projects.
Familiarity with SOC 2 or ISO 27001 from an engineering implementation perspective.
You might work on projects such as:
Build a test harness that injects malicious instructions through prompts, uploaded files, and retrieved documents.
Find and eliminate a path through which one tenant’s information could enter another tenant’s model context.
Introduce capability-scoped credentials so agents receive access only to the data and tools required for a specific task.
Add security regression evaluations for previously discovered AI vulnerabilities.
Build validation and containment around model-generated JSON, XML, queries, scripts, or API payloads.
Prevent untrusted content from triggering sensitive tool calls without deterministic authorization.
Build traceability from user input through model context, output, tool execution, and final side effects.
Detects unusual tool-call sequences, data-access patterns, agent loops, or token consumption.
Threat-model a new agent capability and implement the required controls before launch.
Turn a recurring vulnerability class into a reusable platform guardrail.
In your first three months, you will:
Understand the major trust boundaries across Metaforms’ AI agents, customer data, tools, and infrastructure.
Build an initial threat model and prioritized AI-security testing plan.
Reproduce and help remediate high-priority product or AI vulnerabilities.
Add high-signal security tests for important agent and application boundaries.
Improve visibility into model calls, tool usage, and security-relevant agent behavior.
Over time, you will:
Build continuous adversarial testing into how Metaforms develops AI features.
Reduce recurring vulnerability classes through reusable controls and secure defaults.
Help engineers ship AI capabilities quickly without weakening customer-data boundaries.
Develop into a technical owner for AI security across the platform.
Make Metaforms’ agents more secure, observable, reliable, and resilient as their capabilities grow.
Work on practical AI-security problems in production agent systems.
Operate across offensive testing, application security, and platform engineering.
Ship fixes and controls directly rather than producing reports that sit in a queue.
Work closely with engineering leadership and influence platform design early.
Take meaningful ownership while growing into an emerging technical specialization.
The process will focus on practical engineering judgment rather than security trivia:
Introductory conversation.
Technical discussion covering AI systems, security fundamentals, and past projects.
Practical exercise involving a simplified agent workflow, its trust boundaries, likely attack paths, and one proposed or implemented control.
Final conversation with engineering leadership.
Break AI systems like an attacker. Fix them like a platform engineer.
Similar vacancies
Metaforms
Bangalore, भारत
PRAGMATIKE
Bangalore, भारत
Skyhigh Security
Bangalore, भारत
Syngenta Group
Pune, भारत
Assent
Pune, भारत
Krazybee Services Limited
Bangalore, भारत
L4B Software
Pune, भारत
Hyderabad, भारत
Amtech Software
$100 USD
Bangalore, भारत
EPAM Systems
Pune, भारत