Security Researcher

Engineering

Contract €50k – €100k Remote / In-person San Francisco / Berlin / Remote

We tell companies their secrets never touch the model. That every action is permissioned. That the sandbox actually holds. Your job is to prove us wrong — and then help us fix it. You'll attack ocr from every angle: prompt injection, sandbox escapes, permission bypasses, secret exfiltration. If an agent can do something it shouldn't, you find it first.

We're building a team of geniuses. Not a team of "smart people" — actual geniuses who ship. You've built things before — maybe a company, maybe projects that people actually use. You want to work unreasonably hard on something that matters. If you fall short of that standard, you're wrong here.

• Continuously probe the agent runtime for vulnerabilities — sandbox isolation, permission enforcement, secret handling, integration scoping.
• Develop attack scenarios that model real‑world threats: malicious prompts, tool misuse, multi‑step exploits across agent sessions.
• Write clear, reproducible findings and work with engineering to close gaps.
• Help define security architecture decisions as the platform evolves — threat models, trust boundaries, defense‑in‑depth strategy.

• Deep experience in application security, penetration testing, or red teaming. You've found real bugs in real systems.
• You understand LLM‑specific attack surfaces — prompt injection, jailbreaks, indirect prompt injection, tool‑use exploits. This is not theoretical for you.
• Strong systems background. You're comfortable reading Go, understanding container isolation, and reasoning about permission models at the code level.
• You can communicate findings clearly to engineers who will fix them. No 40‑page reports that sit in a drawer.
• Self‑directed. Contract means you set your own pace, but you deliver consistently and proactively.
• Read our values before applying. We default to open — including about what we get wrong.

• Competitive contract rate.
• You're securing the runtime that companies trust to run AI agents in production. The stakes are real.
• Direct access to the entire codebase and engineering team. No bureaucracy between finding a bug and shipping a fix.
• Early‑stage company where your work directly shapes the security posture of the product.
• Flexible engagement — remote‑first, set your own hours, deliver results.