Principal Researcher, AI Red Teaming (Remote, ITA)

CrowdStrike Holdings, Inc.; Italia; Remote; EUR 106,000 - 163,000; Full time

Benefits offered by this job

Equity awards

Wellness programs

Vacation and holidays

Parental leaves

Professional development

Employee networks

Job summary

CrowdStrike Italy s.r.l. seeks a senior security researcher to lead AI red-teaming efforts, develop exploit research, and publish findings. You will design and run vulnerability assessments of agentic systems, build research tooling, and help shape the external research agenda across venues and disclosures.

The role emphasizes hands-on security research, Python development, and collaboration with legal, product, and engineering teams to communicate risk clearly and drive responsible AI adoption.

Qualifications

• Public record of original security research (CVEs, disclosures, papers)
• Deep hands-on expertise in offensive security: exploitation, vulnerability research, red teaming
• Demonstrated research on LLMs or agentic systems: prompts, tool use, jailbreaks
• Strong Python and systems engineering skills, harnesses and tooling
• Experience directing technical work and guiding others as a tech lead or principal

Responsibilities

• Research and red team frontier models, and publish results with rationale and timelines
• Conduct vulnerability research on agentic ecosystems end-to-end from hypothesis to patch verification
• Design tooling for research: attack harnesses, traces, interceptors, isolated targets
• Produce state-of-the-art research notes, external papers and coordinated disclosures
• Define attack content formats and ensure reproducibility in automated AI security tests

Skills

Python

Systems engineering

Offensive security

LLMs / agentic systems research

Written communication

Technical leadership

Job description

About the Role

As a global leader in cybersecurity, CrowdStrike protects the people, processes and technologies that drive modern organizations. Since 2011, our mission hasn’t changed — we’re here to stop breaches, and we’ve redefined modern security with the world’s most advanced AI-native platform. We work on large scale distributed systems, processing almost 3 trillion events per day and this traffic is growing daily. Our customers span all industries, and they count on CrowdStrike to keep their businesses running, their communities safe and their lives moving forward. We're proud to work for a mission-driven company leveraging AI to transform the way we work. CrowdStrikers drive their careers through flexibility and autonomy while also being expected to contribute to a culture of responsible AI adoption, experimentation, and innovation. We use an AI-first mindset as a force multiplier to proactively and continuously accelerate execution, build expertise, uncover insights, and solve complex problems. We’re always looking to add talented CrowdStrikers to the team who have limitless passion, a relentless focus on innovation and a fanatical commitment to our customers, our community and each other. Ready to join a mission that matters? The future of cybersecurity starts with you.

What You'll Do

Research (about half of your time): Bring the research subjects. Propose and prioritise the attack classes, agent target classes and frontier model behaviours the team studies, with a written rationale for each and a kill date. Red team frontier models and the agentic systems. Track how each model release shifts attack success, agent behaviour and the controls that hold, and publish the result. Conduct vulnerability research and security assessments on agentic ecosystems end to end: hypothesis, exploit, observed action, disclosure, patch verification. Deliver state-of-the-art research: internal research notes on a monthly cadence, external papers, conference talks and coordinated disclosures with vendor credit. Investigate failure modes personally where the evidence is hardest to get: multi-agent systems, long-horizon tasks, defences that look like they work. Tooling for research and red teaming: Design and write the tooling the research depends on: attack harnesses, action-trace capture, tool-call interception, payload converters, isolated lab targets and reproducible experiment runners. Turn each finding into a reusable attack module with an executing payload test in the team's automated AI security testing platform, and set the engineering bar for how that is done. Define the shared formats for attack content and action traces so red team output can be scored, replayed and reused by the benchmarking work without conversion. Technical guidance (no people management): Own the external research agenda for AI red teaming: which venues, which disclosures, which collaborations, and the quality bar for anything that leaves the team. Review benchmark methodology from the other pillar for attacker realism, and accept review of your own findings for evidence quality. Communicate findings, limitations and risk clearly to engineering, product owners, legal and executives. Report what was not tested as plainly as what was.

What You'll Need

A public record of original security research: CVEs, coordinated disclosures, conference talks or peer-reviewed papers that you led. For this level we expect the record to be substantial and verifiable. Deep hands‑on expertise in offensive security: exploitation, vulnerability research or red teaming, with the ability to find and demonstrate a vulnerability end to end yourself. Demonstrated research on LLMs or agentic systems: prompt and environment injection, tool‑use abuse, jailbreaks, multi‑agent attacks, or model evaluation under adversarial conditions. Shareable artifacts required. Strong Python and systems engineering skills. You write the harnesses, interceptors and converters the team runs on, and you review others' code to a production standard. Practical understanding of how agentic systems are built: tool use, planning loops, MCP, memory, sandboxing, and where trust boundaries sit. Experience setting technical direction for other researchers as a tech lead or principal, with evidence that people you worked with went on to lead work themselves. Exceptional written communication. Your research notes, disclosures and paper are the team's external face.

Bonus Points

Published research on AI security, agent security or adversarial machine learning at recognised venues. Experience red teaming frontier models directly, or working with frontier model providers on disclosures. Patents or shipped tooling in automated red teaming, agent evaluation or security testing. Experience with isolated lab design: sandboxing, egress control, C2 sinkholes, telemetry retention.

#LI-JP2 #LI-Remote

Benefits of Working at CrowdStrike

• Market leader in compensation and equity awards
• Comprehensive physical and mental wellness programs
• Competitive vacation and holidays for recharge
• Paid parental and adoption leaves
• Professional development opportunities for all employees regardless of level or role
• Employee Networks, geographic neighborhood groups, and volunteer opportunities to build connections
• Vibrant office culture with world class amenities
• Great Place to Work Certified™ across the globe
• CrowdStrike is proud to be an equal opportunity employer.
• We are committed to fostering a culture of belonging where everyone is valued for who they are and empowered to succeed.
• We support veterans and individuals with disabilities through our affirmative action program.
• CrowdStrike is committed to providing equal employment opportunity for all employees and applicants for employment.
• The pay range for this position for all candidates based in Italy is fixed salary of €106,000 - €162,800 per year base pay.
• The role holder would also be eligible for a comprehensive benefits package (including medical and dental insurance, access to Modern Health wellbeing resources, life and disability coverage).

CrowdStrike Italy s.r.l. is committed to fair and equitable compensation practices. Placement within the stated pay range is dependent on a variety of factors, including, but not limited to, relevant work experience, skills, qualifications, job level, supervisory status, and location.

CrowdStrike was founded in 2011 to fix a fundamental problem: The sophisticated attacks that were forcing the world’s leading businesses into the headlines could not be solved with existing malware-based defenses. Founder George Kurtz realized that a brand new approach was needed — one that combines the most advanced endpoint protection with expert intelligence to pinpoint the adversaries perpetrating the attacks, not just the malware. There’s much more to the story of how Falcon has redefined endpoint protection but there’s only one thing to remember about CrowdStrike: We stop breaches.