AI Security Engineer

Role overview Design and strengthen secure environments for running untrusted AI models and agents. The work combines systems security, monitoring, and adversarial testing to limit what these systems can access, detect harmful activity, and verify that safeguards hold even when a model behaves unexpectedly.

Responsibilities - Harden model inference and agent execution environments by setting process, filesystem, device, and resource limits; assess isolation in sandboxes and virtual machines, including GPU access. - Build infrastructure for offline operation and controlled handling of model weights, dependencies, updates, and exported artifacts. - Keep credentials, policies, and administrative controls outside model access; restrict tools and mediate any approved network actions. - Correlate model outputs and tool calls with process activity, file access, and network attempts, using protected audit trails and local monitoring. - Detect and contain unauthorized access, attempted data exfiltration, privilege escalation, and monitoring interference; preserve evidence and account for detector failures. - Run controlled adversarial evaluations, including prompt-injected agents and compromised artifacts, and measure missed attacks, false alarms, response time, and defense costs.

Requirements - Experience with Linux and infrastructure security, including networking, isolation, least privilege, and secure operations. - Experience building or investigating sandboxes, virtual machines, hardened containers, or comparable environments for untrusted code. - Ability to build security tools and automation in Python, Go, Rust, or a similar language, and work with model-serving or agent runtimes. - Background in detection engineering, incident response, or adversarial testing, with an understanding of false positives and monitoring limits. - Skills in threat modeling, reproducible security testing, and clear technical writing.

Nice to have Relevant experience may come from systems security, malware analysis, infrastructure, or security research. A previous AI security job title is not required.

Benefits and work setup The role is remote. Applicants are asked to provide a CV or professional profile and an example of relevant work, such as a repository, security tool, threat model, or technical investigation. For confidential work, a non-sensitive summary of the problem, approach, and findings is acceptable.