AI Security Researcher

Job description

## About Cogent Cogent is an Applied AI Lab building the next generation of AI agents for cybersecurity. AI has fundamentally changed how attacks happen, allowing malicious actors to operate at unprecedented speed and scale. Cogent's "AI Taskforce" assesses petabytes of enterprise data to remediate these issues before critical breaches occur. To stay at the cutting edge, we blend frontier research with real-world execution. Alongside our core product work, Cogent Research serves as our applied AI lab, providing the research horsepower needed to make truly agentic security workflows a reality. We partner with Fortune 500 companies to secure some of the most complex production environments in the world. We're backed by Greylock & Bain Capital Ventures, and we've built a team with the best minds in applied AI. Our team is comprised of people from: - Top universities like Stanford, Berkeley, Penn, Duke, Carnegie Mellon, Waterloo - Unicorn, high-growth companies like Scale AI, Databricks, Stripe, Tesla, Coinbase - World class cybersecurity experts from Wiz, Abnormal AI, Zscaler - Preeminent research labs like Deepmind and SAIL ## About the Role We're hiring a Security Researcher to build the offensive side of our research - the attacker our defensive agents have to beat. You'll sit in Research and own the loop that turns real red team tradecraft into AI capability, shaping how offensive expertise becomes autonomous capability rather than just supplying the expertise. ## What You'll Do - Design and build our red team agents. Architect the agent harness, the tools and Model Context Protocol (MCP) servers they call, and the skills that encode how an operator actually works - from scoping and reconnaissance through payload development, delivery, Command and Control (for example Mythic), and post-exploitation toward the objective - turned into repeatable, agent-executable capabilities for our offensive agent and VR-1. You'll review the agents' runs and close the gaps. - Push the agents to full-chain operations across every surface. Our red team agent covers application and code security today. You'll drive its extension into full attack chains - initial access, delivery and evasion, Command and Control, privilege escalation, and lateral movement - across Windows, Active Directory, and on-prem, and work with our cloud security researchers to extend coverage into cloud (identity and access management, misconfiguration, continuous integration and delivery) and network. - Own the evasion and tradecraft edge. Build detection-evasion know-how into the agents and test harness, and the detonation and detection loop that validates it against real Endpoint Detection and Response (EDR). This is what lets our defensive agent train against a real adversary instead of a toy. - Design and operate the cyber ranges this runs in. Build realistic environments with real defenses in place (EDR, Security Information and Event Management, logging), stood up in-house and via vendors like Ludus - Windows and Active Directory-heavy to start, extending to cloud and hybrid alongside the cloud security team. These are where agents are trained, benchmarked, and verified. - Build the benchmarks that prove the agents improved. Turn attack paths into evaluation tasks with verifiable success criteria, feeding benchmarks like IntrusionBench, so we can measure whether each change actually makes the agents better. - Verify critical and high-severity attack paths in real customer environments, and feed what you learn back into the agents and detections. - Represent Cogent in the offensive security community through research, talks, and open-source contributions. ## What We're Looking For - 6+ years of experience in offensive security, red team operations, or adversary emulation - Hands-on operator who has run full-chain operations end-to-end - initial access through post-exploitation - across Windows, Active Directory, and on-prem environments - Depth in Command and Control frameworks (for example Mythic) and payload development, with real detection-evasion experience against EDR - A builder who writes code (for example Python) to turn manual tradecraft into repeatable, automatable capability rather than one-off operations - Experience standing up and operating realistic test environments or cyber ranges with production defenses in place (EDR, SIEM, logging) ## Bonus Points - Cloud (identity and access management, misconfiguration, CI/CD) or network attack experience - Experience building evaluations or benchmarks that measure security capability - Open-source tooling, conference talks, or published research in offensive security - Experience working with AI agents, LLM tooling, or agentic workflows