AI Security Engineer.
Remote
Build secure environments for untrusted AI models and agents.
The role focuses on containing harmful actions, including behavior originating from the model itself. You’ll combine systems security, observability, and adversarial testing to establish what a model can access and verify that those boundaries hold.
How to applyThe role
Model containment
Harden inference and agent-execution environments. Define process, filesystem, device, and resource limits, and evaluate isolation in sandboxes and virtual machines, including GPU access.
Air-gapped infrastructure
Build systems that operate without external network connectivity. Establish controlled procedures for importing model weights and dependencies, applying updates, and reviewing exported artifacts.
Permissions & secrets
Keep credentials, policy, and administration outside the model’s control. Restrict tool access. For deployments that require connectivity, deny network access by default and mediate approved actions.
Behavioral observability
Connect model outputs and tool calls to process activity, file access, and network attempts. Build protected audit trails and local monitoring that can reconstruct a suspicious run independently of the model.
Detection & response
Detect unauthorized access, exfiltration attempts, privilege escalation, and monitoring interference. Block actions at execution boundaries, stop or quarantine runs, and preserve evidence. Plan for detector failures.
Adversarial evaluation
Test containment and monitor evasion in controlled environments, including prompt-injected agents and compromised artifacts. Measure missed attacks, false alarms, response time, and the cost of defenses.
Your experience
- Linux and infrastructure security, including networking, isolation, least privilege, and secure operations.
- Building or investigating sandboxes, virtual machines, hardened containers, or other environments for untrusted code.
- Security tooling and automation in Python, Go, Rust, or comparable languages, with the ability to work with model-serving and agent runtimes.
- Detection engineering, incident response, or adversarial testing, including an understanding of false positives and monitoring limitations.
- Threat modeling, reproducible security tests, and clear technical writing.
Relevant backgrounds include systems security, malware analysis, infrastructure, and security research. A previous AI security job title is not required.
How to apply
Email hr@ovrlab.io with the subject AI Security Engineer — Application — Your Name, replacing “Your Name”. Include:
- A short introduction and your CV or professional profile.
- An example of relevant work: a repository, security tool, threat model, or technical investigation. Explain your contribution and findings.
You can submit existing work. If it’s confidential, send a summary of the problem, your approach, and the outcome, omitting sensitive details.
Our privacy policy explains how we handle application correspondence.