We are building observability and security software that will let anyone ask their OS what occurred in app space, generate human readable reports of app and agent behavior, and build in security they can understand. We are researching a variety of novel approaches to agent detection, agent intent inference, runtime security, and hierarchical summarisation of OS-level telemetry.
The tech stacks listed under each role are estimates of what is likely to be used. C, Rust, and PyTorch are non-negotiable for the parts of the system that need them. Everything in between is flexible.
Research positions prefer candidates with advanced degrees in quantitative fields; advanced-degree dropouts are welcome (our founders are a Systems Engineering MS dropout and an ML PhD dropout). Engineering positions prefer candidates with green GitHub accounts and project work that demonstrates knowledge of the key areas in the role description.
If you have ever been blocked from shipping a solution you knew was superior, prevented from refactoring detrimental code for bureaucratic reasons, or hindered in any way from delivering your best software; that will not happen to you here.
Roles are based in London or San Francisco. Remote is possible at this early stage but it is likely that we will migrate to in person only at later stages. Compensation is competitive and adjusted for region.
Open Roles
This role is for those with an intimate understanding of OS abstractions. You will work on telemetry gathering, state management, high-performance stream processing, and optimising machine learning inference for edge devices. You might be a good fit if OS abstractions and implementation details are something you study and implement for fun.
SkillsC · Rust · eBPF · Linux kernel internals · Linux Security Module · macOS Endpoint Security Framework · Seatbelt · Bubblewrap · concurrency primitives and lock-free data structures · SIMD and threading models · high-throughput streaming systems · compute and inference optimisation · perf tooling
This role researches approaches to agent detection, agent intentionality inference, language modelling for hierarchical summarisation of OS-level telemetry, and mathematical extensions to contemporary runtime security. Candidates will typically hold a PhD, or be enrolled in a graduate program in Physics, Mathematics, Statistics, or another quantitative field. You might be a good fit if you have to hold yourself back from dropping Information Theory, Point Process Aware Time Series, Digital Signal Processing, Symbolic Dynamics, or Automata Theory into casual conversation.
SkillsPyTorch and any data analytical stack you prefer
This role builds scalable machine learning pipelines, reinforcement learning environments, custom agents in various configurations, software that replicates distributed deployments, and anything else our researchers need to create and scale their experiments and synthetic data generation runs. You might be a good fit if you have built secure and scalable data pipelines and can anticipate the needs of researchers.
SkillsPython · Rust · DuckDB · BigQuery · Dataflow (Apache Beam) · Pub/Sub · Vertex AI · Vertex AI Pipelines (Kubeflow) · Cloud Composer (Airflow) · Ray · GKE · Terraform · MLflow or Weights & Biases
Adversarial evaluation of our EDR features. You will attempt to bypass agent detection, evade behavioural classification, and exfiltrate data through agents in ways our models miss, then write up what worked so we can close the gap. This role defines the bar for what our detection actually catches. You might be a good fit if reverse-engineering detection systems and finding evasion paths is something you do for fun.
SkillsRed-team operations · EDR evasion techniques · malware behavioural analysis · syscall-level exploitation · threat modelling · agent-as-attack-surface methodology
If any of these roles sounds like you, contact careers@perpetualautomata.com with your resume and a link to your GitHub.