Staff+ Software Engineer, Anthropic

• 10 + years experience designing, building, and operating detection, anti-fraud, anti-abuse, or security systems in production
• A track record of independently scoping and delivering complex, ambiguous, multi-month technical projects
• Experience making architectural decisions in an adversarial domain that other engineers and teams then build on
• Proficiency in Python and SQL, with strong software engineering fundamentals and hands-on coding ability
• Experience leading investigations into account-based abuse or security incidents, and translating findings into automated detection
• Ability to reason rigorously about large behavioural or telemetry datasets, and to distinguish attacker behaviour from unusual but legitimate use
• Strong written communication and a track record of driving alignment across multiple teams and stakeholders
• Sound judgement about the tradeoff between stopping bad actors and disrupting legitimate users, and the ability to explain and defend where you have drawn that line
• Minimum education: Bachelor’s degree or an equivalent combination of education, training, and/or experience
• Required field of study: A field relevant to the role as demonstrated through coursework, training, or professional experience
• We encourage you to apply even if you do not believe you meet every single qualification

Desirable

• Anthropic's Safeguards organization builds the systems that keep Claude safe to use at scale. The Account Compromise team owns one specific slice of that work: protecting the people and organizations who use Claude from losing control of their accounts, and preventing compromised accounts and credentials from being used to abuse our platform.
• Account takeover, credential stuffing, phishing-driven session theft, leaked API keys, and resold access all cause real harm — to the customers whose accounts are hijacked, and to the wider ecosystem when stolen access is used to route abusive traffic through legitimate accounts
• We are looking for a staff+ engineer to set the technical direction for this work. You will own the architecture of how we detect, contain, and remediate account compromise across Claude and the Claude Developer Platform, making the design decisions that other engineers and teams build on top of
• You will move between deep technical work and adversarial problem-solving: threat modelling how attackers will adapt, scoping multi-month projects from ambiguous starting points, and leading complex live investigations when they escalate
• This is a hard problem space. Attackers iterate quickly, the signals separating a compromised account from an unusual but legitimate one are subtle, and every defence carries a cost to real users if it fires incorrectly. You will be building the playbook, and the judgement calls about where to draw those lines will largely be yours to make and defend
• You will operate with high autonomy — owning detection coverage and incident leadership, driving alignment with Security, Product, and Policy teams, and shaping how Anthropic approaches this problem globally rather than executing against someone else's roadmap
• Set the technical direction and own the architecture for account compromise detection, response, and remediation across Claude and the Claude Developer Platform
• Independently scope and lead complex, multi-month engineering projects from an ambiguous starting point through to production systems that operate reliably under adversarial pressure
• Build and evolve detection systems that identify account takeover, credential abuse, and compromised API keys in near real time
• Design automated response flows that cut off attacker access while minimising disruption to legitimate users
• Lead investigations into significant compromise incidents end to end, then convert what you learn into durable, automated defences
• Threat model how attackers are likely to adapt, and prioritise the team's work against that view rather than only against incidents already observed
• Drive cross-organisational alignment on account security direction with Security, Product, Support, Policy and other partners
• Define how the team measures success and hold the work to those measures
• Set technical standards for the domain and raise the bar for other engineers through code review, design review, and mentorship
• Surface patterns from compromise cases to research and product teams so that protections improve upstream
• Representative projects
• Design the architecture for real-time login risk scoring, and get agreement across Security, Product, and Safeguards on where step-up authentication should and should not fire
• Design detection for compromised API keys, together with security controls to limit exposure
• Rebuild product features to regain access quickly when customers get compromised
• Write the threat model that sets the team's roadmap for the next year, and bring the rest of the organisation along with it
• Instrument a false positive review loop that measures how often the team's defences affect legitimate users, and drive that number down