Staff Security Researcher, Offensive AI

The Browser Company

8d • $225,000 - $300,000 • Remote

About The Position

Dia is a browser enhanced with agentic capabilities. The agent reasons over untrusted content from the open web and takes real actions on the user's behalf, inside a client that sits next to everything the user is signed into. That combination produces a threat model that mostly doesn't have prior art — the interesting bugs aren't on a checklist, and the tooling to find them largely doesn't exist yet.
As a Staff Security Researcher on our security team, you'll do original offensive research against Dia including the client, agent runtime, tools and integrations it calls, and services behind them. You'll work ahead of the product, threat modeling new surfaces with the teams designing them, and reviewing features before they reach users.
You'll also eliminate bug classes by building model-driven scanning, fuzzing, and agentic hunting systems that run continuously against our code, our infrastructure, and our agent.
You’ll partner closely with the engineers who own remediation, rather than owning the fixes yourself. You'll report to the Head of Security and work across client, infrastructure, and product engineering.

Requirements

• 8+ years in offensive security — vulnerability research, exploit development, red teaming, or product security testing — with a record of finding real bugs in software other people had already reviewed.
• Depth in at least one hard surface: LLM agent systems, browser or Chromium internals, OS sandboxing and native clients, or backend and cloud infrastructure — and the excitement to learn the rest.
• Practical fluency using LLMs as instruments, not just as targets, plus a working sense of when their output is noise.
• You write production-quality code in one or more of Go, TypeScript, Python, or Swift. You'd rather build the thing that finds a bug a thousand times than find it once.
• You write findings engineers act on, and you can stay in the fix conversation and push for the durable version without owning the remediation yourself.
• You thrive in a high-trust, high-ambiguity environment: you seek feedback, but you don't need hand-holding.
• You resonate with our company values.
• We're primarily focused on hiring in North American time zones and require that folks have 4+ hours of overlap time with team members in Eastern Time Zone.

Nice To Haves

• Continuous AI-assisted vulnerability discovery (first project). You’ll survey commercial scanning services, frontier models, open-source security models, and stand up a working mix against our codebase, our infrastructure, and the agent runtime.
• Agent red teaming. Plan and run regular AI-assisted red team exercises against Dia and our infrastructure, and build the attack corpus and harness that make each round cheaper than the last.
• Pre-launch review as a practice. New tools and integrations, agent capabilities, enterprise controls, platform expansion. Enhanced with platforms to assist with testing
• Structural fixes that eliminate classes of vulnerability. Define, with the product engineers, the changes that take a whole bug class off the table instead of fixing the same issue one report at a time.
• Future tools and systems. The state of the art in vulnerability hunting is evolving faster than it ever has. We expect this person to track it and keep BCNY best in class, which means therea re opportunities to build far beyond the scope described above.

Responsibilities

• Run original offensive research against Dia, its agent, and backend services, focusing on prompt injection and indirect exfiltration, tool-call abuse, permission and provenance bypass, sandbox escape, cross-profile data access, quota and abuse-scoring bypass.
• Threat model new surface areas with the teams building them, and review features before they ship, identifying what is exploitable, what it costs an attacker, and what would have to be true to launch.
• Design and build automated vulnerability discovery including model-driven code and infrastructure scanning, fuzzing harnesses, and agentic hunting pipelines that keep working after the engagement ends.
• Eliminate entire classes of bugs in collaboration with the engineers who own the fix, then make the same issue structurally hard to reintroduce through an enforced invariant, a fail-closed default, a test, a lint, or a platform change.
• Set the standard for what "security tested" means before a feature ships, and raise the ceiling on what the whole team can find.

Benefits

• base salary
• equity
• comprehensive benefits
• best-in-class benefits designed to support you, your family, and your life outside of work
• big-company perks with startup-style impact, ownership, and ways of working

What This Job Offers

• Job Type: Full-time
• Career Level: Senior
• Education Level: No Education Listed