About the role:

Anthropic's Application Security team secures the systems that build, serve, and increasingly are Claude. The attack surface is unlike most AppSec work: multi-agent orchestration, sandboxed code execution, agents holding delegated credentials, untrusted tool output crossing trust boundaries — problems with little prior art and no off-the-shelf playbook.

The way the team works is also different. We use Claude as our primary tool across every part of the job: it drives our static analysis, drafts and fixes vulnerabilities as pull requests, performs first-line bug bounty triage, and assists threat modeling for design reviews. The human work is the judgment layer — system-level reasoning, deciding what matters, and building the next thing the model can't do yet.

This is a builder's role on a senior team. We hire engineers who ship production systems and clear a hands-on threat-modeling bar — people who can find the vulnerability but would rather build the system that finds them all. Every engineer owns a system end-to-end, and the team's work has shaped customer-facing product security — including Claude Code's security review tooling, its security guidance plugin, sandboxing, and auto mode.

Key responsibilities:

• Design, build, and operate Claude-powered security systems — LLM-driven code analysis, automated vulnerability remediation, AI-assisted threat modeling — and own one or more of them end-to-end, including the cross-functional relationships that come with it
• Lead secure design reviews and threat modeling for novel AI systems, identifying risks that don't map to existing frameworks
• Evolve a public bug bounty program where automation handles routine triage and root-cause work, and engineers handle escalations and corner cases
• Partner with Product, Infrastructure, and Research teams as an embedded security owner — consulting on launches, shaping architecture, and influencing decisions where security is the constraint
• Share an operational on-run rotation with the rest of the team — bounty escalations, incident response, and launch consults on systems serving Claude in production

Minimum qualifications:

• Hands-on application and infrastructure security experience, including cloud and containerized environments
• Production-quality coding ability in at least one of Python, Go, Rust, or TypeScript, with a track record of building durable systems rather than one-off scripts
• Practical threat-modeling and vulnerability-identification skills — you've found and reasoned about real bugs in real systems, even if breaking isn't your primary mode
• Demonstrated ability to operate with high autonomy and ambiguity — comfortable being handed a problem and a lot of latitude rather than a spec
• Clear technical communication with both engineers and leadership

Preferred qualifications:

• 7+ years in application security, security engineering, or security-focused software engineering
• Already use LLMs as a core part of how you work, with opinions about where they help and where they don't
• Experience securing agentic, code-execution, or LLM-integrated systems specifically
• Prior ownership of a bug bounty program, vulnerability disclosure program, or vulnerability-management infrastructure at scale
• Background building security automation or developer-facing security tooling
• Offensive security or penetration testing experience

Representative projects

• An increasingly autonomous vulnerability pipeline — LLM-driven code analysis finds the issue, scores it for real exploitability, and opens the fix PR, with humans as the review step rather than the author
• Bug bounty operations where Claude handles first-line triage and drafting, and engineers focus on the reports that actually need judgment
• AI-assisted threat modeling that generates intake questions, drafts the model, and recommends which design reviews need a human in the room
• Automated dependency vulnerability remediation across Anthropic's codebase
• The company-wide vulnerability dashboard and SLA enforcement layer every engineering team works against
• Threat models and security architecture for agentic product surfaces — code execution sandboxing, agent identity and delegated auth, tool-use boundaries

Annual Salary:

$320,000 - $485,000 USD