Security Engineer, Agent Security

Plasma AI; New York (NY); On-site; USD 120,000 - 160,000; Full time

Job summary

Plasma is seeking a security engineer in New York to define and enforce boundaries for agent behavior. You will build controls, monitoring, and response mechanisms to keep agents within user-defined limits while they operate unattended for long periods.

The role blends research and product work: study how agents can be misused and ship defenses that integrate with Plasma's platform, identity, and isolation layers.

Qualifications

• Experience in application, infrastructure, or offensive security, with a record of shipping secure production services.
• Strong programming skills in Python or a systems language such as Go or Rust
• Experience with isolation techniques such as containers and sandboxing, and with identity-based access control
• Cloud security experience on at least one major provider
• Familiarity with attacks specific to LLMs, such as prompt injection, is a plus

Responsibilities

• Threat modeling: Map how agents can be attacked or misused, including prompt injection, data exfiltration, and privilege escalation between agents
• Runtime controls: Build the policy enforcement that holds agents to their permissions and budgets, on top of the platform's identity and isolation layers
• Red teaming: Attack Plasma's own systems and fix what you find
• Detection & Response: Build monitoring that flags agents acting outside their limits, and the response process that follows
• Customer Security: Work with customers' security teams on their reviews, and design Plasma's controls to meet their requirements

Skills

Threat modeling

Runtime controls

Red teaming

Detection

Response

Customer security

Python

Go

Rust

Containers

Sandboxing

Identity access

Cloud security

LLMs security

Tools

AWS

Job description

We're looking for a security engineer to make sure agents on Plasma stay within the limits users set. You'll build the controls, monitoring, and response that hold those boundaries while agents do consequential work unattended, for hours at a time.

You've secured production systems and want to work on one that existing playbooks don't cover. The role is both research and product: agent security is a young field, so you'll study how agents get misused and ship the defenses into Plasma.

What You'll Work On

• Threat Modeling: Map how agents can be attacked or misused, including prompt injection, tool misuse, data exfiltration, and privilege escalation between agents
• Runtime Controls: Build the policy enforcement that holds agents to their permissions and budgets, on top of the platform's identity and isolation layers
• Red Teaming: Attack Plasma's own systems and fix what you find
• Detection & Response: Build monitoring that flags agents acting outside their limits, and the response process that follows
• Customer Security: Work with customers' security teams on their reviews, and design Plasma's controls to meet their requirements

What We're Looking For

• Experience in application, infrastructure, or offensive security, with a record of shipping secure production services
• Strong programming skills in Python or a systems language such as Go or Rust
• Experience with isolation techniques such as containers and sandboxing, and with identity-based access control
• Cloud security experience on at least one major provider
• Familiarity with attacks specific to LLMs, such as prompt injection, is a plus