Staff AI Security Researcher

The Opportunity: Staff AI Security Researcher (Singapore Based)

We are hiring a Staff AI Security Researcher with a strong offensive security mindset, published CVEs, and a proven ability to use AI as a hunting tool. This role finds real, exploitable issues before attackers do, with external (internet-facing) attack paths first and internal ones next.

You will set the AI security research agenda, build autonomous agents that hunt and iterate on the fly, and jailbreak AI systems. You will also build the workflows, quality bar, and controls that the rest of the company uses.

The ideal candidate has a deep offensive security background, has already shifted day-to-day work onto AI-assisted hunting, and is comfortable leading that shift across teams.

Key Responsibilities

• Set the AI security research agenda: decide which external attack paths, products, and agent or MCP surfaces matter, and why.
• Proactively find, exploit, and document real-world vulnerabilities, prioritizing external (internet-facing) attack paths, then internal ones.
• Discover these issues before attackers do, with working proofs of concept and demonstrated impact.
• Own findings from proof of concept through engineering fix, verification, and the control that stops the same class of issue next time.
• Build autonomous agents that hunt, attack, and iterate on the fly by interacting with AI models and tools.
• Build the hunting system other people run: playbooks, agent workflows, tool-calling setups, and a repeatable bar for proof of concept and demonstrated impact.
• Use and develop AI-driven offensive workflows (multi-model setups, tool-calling, notebooks, orchestration) for continuous hunting.
• Apply deep offensive security skills (API abuse, authentication and access control, secrets, misconfiguration, red teaming) with AI in the loop.
• Produce CVE-quality research outcomes, including work that changes how a real product ships.
• Raise the researcher bar: review other people’s findings, hold a high standard for exploitability, impact, and write-up quality, and teach that standard.
• Design, execute, and document jailbreaks, prompt injection attacks, model evasion, data exfiltration, and other offensive techniques against AI systems.
• Assess and attempt to compromise Model Context Protocol (MCP)–based systems and other tool-calling / plugin ecosystems.
• Take on company-level problems that do not yet have a ticket: third-party model and plugin risk, MCP and tool ecosystems, agent supply chain, and data exfiltration paths across products.
• Build and automate security testing workflows involving multiple models, APIs, and tools (e.g., Jupyter notebooks, orchestration frameworks).
• Perform offensive security testing and red teaming of AI-driven products, including API manipulation and integration abuse.
• Research security weaknesses in models and the infrastructure around them.
• Own the guardrail strategy for models and AI automations: define scope, implement and test safety and security guardrails, and check that they hold in production.
• Propose and evaluate defensive controls: input/output filtering, policy enforcement, monitoring, anomaly detection, and non-AI controls to secure AI systems.
• Define how the program is measured — external issues found before attackers, time to a verified fix, repeat-class rate, and guardrail bypass rate — and report that to engineering leadership.
• Translate research findings into practical engineering requirements and collaborate closely with product and engineering teams to implement fixes and mitigations.
• Influence teams you do not belong to. Explain risk and impact clearly, and turn findings into priority, funding, or a shipped control with product, engineering, and C-level stakeholders.
• Produce clear technical documentation, proofs of concept, and knowledge sharing on AI security practices and new attack and defense techniques.
• Represent the work beyond the team through original research, internal standards, and clear briefings when the risk is company-wide.

What you'll Need to Succeed:

Must have

• Bachelor's in Computer Science or related degree.
• 7–12 years in offensive cybersecurity, with work whose impact outlived the individual finding.
• Published CVE(s), including at least one that led to a vendor fix, a control, or a concrete change in how a product shipped.
• Strong offensive security knowledge: API security testing and manipulation; prior red teaming, penetration testing, or adversarial testing.
• Hands-on jailbreak experience (prompt injection, role confusion, data leakage, safety bypasses, and similar).
• Already using AI agents as the daily way to find and exploit bugs, with write-ups or proofs of concept that show the model did real work.
• Experience building an autonomous or semi-autonomous hunting loop (multi-model, tool-calling, orchestration).
• Ability to work with multiple teams, explain risk and impact, and present to different teams and C-level stakeholders.
• Good communication skills in English.
• Experience handling and maintaining production-level code.

Strongly recommended

• Bug bounty / HackerOne or similar track record, as evidence of external exploit judgment.
• Conference presentations or public original research.
• Hands-on work against MCP