How to apply for Red Team Engineer, Safeguards

About Anthropic

Anthropic is a frontier AI research and product company working on alignment, policy, and security. It posts specific high-impact opportunities and explicitly notes that it does not necessarily recommend working at other positions at the company. Candidates should read the linked 80,000 Hours career review about doing harm by working at a frontier AI lab before applying.

About the role

This Red Team Engineer role sits on Anthropic's Safeguards team and focuses on adversarial testing of deployed AI systems and products. The work spans technical infrastructure vulnerabilities and emergent risks from advanced AI capabilities, including coordinated account manipulation, payment fraud, and novel exploitation of product features. You will simulate sophisticated threat actors who chain multiple attack vectors.

A typical day

The job post does not describe a daily routine. Expect to ask about team structure, how testing is scoped, and how findings are reported during interviews. The post does indicate the work involves adversarial testing across product surfaces and researching new approaches for emerging AI capabilities.

Who Anthropic is looking for

• Has hands-on red team or offensive security experience, including chaining multiple exploitation techniques rather than testing single vectors in isolation
• Can move between traditional security work (infrastructure, accounts, payments) and novel AI-specific abuse cases
• Is comfortable designing creative attack scenarios against product surfaces and documenting them clearly
• Can research and implement new testing approaches for emerging AI capabilities as they appear

Tips for this application

• Read the 80,000 Hours career review linked in the job post and address your reasoning about working at a frontier AI lab somewhere in your application.
• Describe specific attack chains you have built, not just individual vulnerabilities you found. The role emphasizes chaining multiple vectors.
• Show examples of adversarial work against product surfaces (accounts, payments, feature abuse), not only network or infrastructure pentesting.
• Note that this is a remote US role on the Safeguards team; state your US work eligibility and remote setup clearly.
• Because Anthropic posts specific opportunities and does not necessarily endorse other roles there, apply to this exact posting rather than a general application.

What to cover in your cover letter

['Concrete red team engagements where you simulated a sophisticated threat actor and chained multiple attack vectors.', 'Experience with abuse cases like coordinated account manipulation or payment fraud, and how you tested for them.', 'Any work on AI systems or AI-adjacent products, including novel abuse specific to model capabilities.', 'Your reasoning about working at a frontier AI company, referencing the concerns Anthropic itself links to.']

Research before applying

• Read Anthropic's mission statement and its published work on alignment, interpretability, and steerability.
• Read the 80,000 Hours career review on working at an AI lab, which Anthropic links in the job post.
• Look at Anthropic's product surfaces (such as Claude) and think about where abuse or fraud could occur.
• Check what Anthropic has published about its Safeguards team and red teaming practices.

Likely interview topics

Based on the job description, expect questions about:

• Walk through a red team engagement where you chained multiple attack vectors to reach an objective.
• How would you approach testing an AI product surface for abuse that has no direct precedent in traditional security?
• How do you prioritize between infrastructure vulnerabilities and emergent AI capability risks?
• Describe a time you found a vulnerability before malicious actors could exploit it. What was your process?
• What is your view on the risks of working at a frontier AI lab, and how do you weigh them?

Common mistakes to avoid

• Treating this as a standard security engineering role and ignoring the AI-specific abuse cases the post emphasizes.
• Applying without engaging with the ethical concerns Anthropic itself raises about working at a frontier AI lab.
• Listing tools and certifications without showing specific attack scenarios you designed or executed.