Kỹ sư Red Team, Safeguards tại Anthropic
Đã xác nhận còn mở
Kiểm thử sản phẩm AI của Anthropic để phát hiện jailbreak, prompt injection và đường lạm dụng khác; xây framework red team tự động cho agent và bảo mật sản phẩm.
Tìm việc theo công ty: Anthropic
- Địa điểm
- Từ xa
- Khu vực tuyển dụng
- Làm việc từ xa-Friendly (Travel Required) | San Francisco, CA
- Hình thức làm việc
- remote
Lưu việc làm — tài khoản miễn phí Việc làm đã lưu
Tạo hồ sơ ứng tuyển miễn phí — Lưu kỹ năng và dự án trong một hồ sơ có thể dùng lại. Hồ sơ mặc định riêng tư; bạn quyết định khi nào chia sẻ. Ứng tuyển tin này trực tiếp với nhà tuyển dụng.
Bằng cách tiếp tục, bạn đồng ý với cách xử lý dữ liệu để quản lý tài khoản và hồ sơ của mình. Quyền riêng tư
Tổng quan vị trí
Nhóm Safeguards của Anthropic tìm Kỹ sư Red Team để phát hiện điểm yếu trong sản phẩm AI đã triển khai trước khi kẻ tấn công khai thác. Phạm vi gồm hạ tầng sản phẩm, hành vi mô hình và các kiểu lạm dụng như thao túng tài khoản có tổ chức hoặc gian lận thanh toán.
Trách nhiệm
- Thực hiện kiểm thử đối kháng trên các bề mặt sản phẩm, kết hợp nhiều kỹ thuật khai thác và mô phỏng chuỗi tấn công thực tế.
- Phát triển phương pháp kiểm thử cho hệ thống agent, việc dùng công cụ và các cách tương tác AI mới.
- Xây framework tự động để liên tục đánh giá bảo mật và nguy cơ lạm dụng.
- Phối hợp với nhóm sản phẩm, kỹ thuật và chính sách để chuyển phát hiện thành biện pháp bảo vệ và chỉ số phát hiện.
Yêu cầu
- Có kinh nghiệm kiểm thử xâm nhập, red team hoặc bảo mật ứng dụng.
- Có kinh nghiệm kiểm thử jailbreak mô hình và luồng agent để tìm đường prompt injection.
- Có kỹ năng bảo mật web thực hành và sử dụng công cụ như Burp Suite, Metasploit hoặc script tự viết.
- Có kinh nghiệm tạo công cụ kiểm thử bảo mật tự động hoặc framework kiểm thử riêng cho LLM, và trình bày rõ kết quả.
Công nghệ trong công việc
- LLM: Experience creating security automation or LLM-specific test frameworks and communicating findings .
- CAN: In this role, you'll take an adversarial approach to uncover vulnerabilities across our product ecosystem before they can be exploited by malicious .
Hướng dẫn đánh giá mức độ phù hợp (tiếng Anh)
Phân tích dựa trên yêu cầu và trách nhiệm trong tin tuyển dụng của nhà tuyển dụng.
Understanding the scope. This Red Team Engineer, Safeguards position at Anthropic is best assessed through the work described in the employer’s posting, rather than the title alone. The available role overview says: “Anthropic’s Safeguards team seeks a Red Team Engineer to find weaknesses in deployed AI products before attackers exploit them. The work spans product infrastructure, model behavior and abuse patterns such as…” Read that statement together with the responsibilities below to understand the likely balance between delivery, coordination and technical decision-making. The role belongs to the technical area, but that label is only a guide to the work; the employer’s own requirements remain decisive. Before applying, check the original listing for reporting lines, product or team assignment, employment terms and any details that may have changed since this copy was checked. Where this page says a fact is not specified, treat it as an open question rather than an implied promise.
What the responsibilities signal. The posting specifically names “Run adversarial tests across product surfaces, combining multiple exploitation techniques and simulating realistic attack chains.”; “Develop testing methods for agent systems, tool use and other emerging AI interactions.”; “Build automated frameworks for continuous security and abuse assessment.”. For each duty, identify a project where you performed comparable work and be ready to describe the context, your own contribution, constraints, decisions, and measurable result. Explain how you moved from the initial need to implementation, review, release or operational follow-through, where those stages apply. If the work involved other teams or users, distinguish your decisions from the group’s contribution. Avoid presenting a technology keyword as proof by itself: connect it to the problem you solved, the scale or complexity involved, and what changed after your work. The source may contain additional duties; review it directly before preparing your application.
How to read the requirements. The listed qualifications include “Experience in penetration testing, red teaming or application security.”; “Experience testing model jailbreaks and agent workflows for prompt-injection paths.”; “Practical web security skills and tools such as Burp Suite, Metasploit or custom scripts.”. Mark each requirement as demonstrated, adjacent experience or a genuine gap, then use a concise example for the first two categories. A strong example explains what you personally owned, how you made trade-offs, how you validated the result and what you learned. If the posting specifies years of experience, language ability, certifications, location or eligibility, address those facts plainly in your CV or application rather than relying on the employer to infer them. Do not claim a qualification from exposure alone. For a requirement you do not meet, describe relevant transferable experience accurately and ask the recruiter whether it is mandatory.
Technology and working evidence. The technology terms associated with this listing are LLM: Experience creating security automation or LLM-specific test frameworks and communicating findings ., CAN: In this role, you'll take an adversarial approach to uncover vulnerabilities across our product ecosystem before they can be exploited by malicious .. Treat this as a set of posting-derived signals, not a guarantee that every tool is used daily or that the list is complete. For each term relevant to your experience, prepare one concrete example: the system or workflow, the part you owned, the alternatives considered, and how you tested reliability, security, performance or usability as appropriate. If you have not used a named tool, explain the closest related system you have worked with and how you would ramp up; do not relabel that experience as direct production use. If the source describes technologies beyond this list, prioritize its exact wording.
Evidence to put in the application. Tailor the first half of your.
Phúc lợi
- Tin tuyển dụng nêu chế độ nghỉ phép, nghỉ chăm con, giờ làm linh hoạt và chương trình đối ứng tùy chọn khi quyên góp cổ phần.
- Anthropic cho biết nhân viên thường làm việc tại văn phòng ít nhất 25% thời gian; một số vị trí yêu cầu nhiều hơn. Hãy kiểm tra điều kiện địa điểm và đi lại của vị trí này trong tin gốc.
Về Anthropic
Anthropic phát triển hệ thống AI, chú trọng độ tin cậy, khả năng diễn giải, khả năng điều khiển và an toàn.
Trang web nhà tuyển dụngCách ứng tuyển
Kiểm tra tin gốc và gửi hồ sơ trực tiếp qua trang tuyển dụng của nhà tuyển dụng.
Xem tin gốc và ứng tuyểnNguồn được kiểm tra: 2026-10-20
Việc làm liên quan
- Kỹ sư bảo mật trưởng nhóm Red Team và tình báo mối đe dọa tại Grab — Grab
- Kỹ sư backend cấp cao cho API Engine tại Deepgram — Deepgram
- Kỹ sư phần mềm tìm kiếm P2P tại Tether — Tether
- Trưởng nhóm Kiến trúc sư Giải pháp Khách hàng tại Supabase — Supabase
- Kỹ sư phần mềm backend cho tác nhân giọng nói tại Deepgram — Deepgram
- Member of Technical Staff, Integration/RL Team (Kỹ sư Nghiên cứu) tại Cohere — Cohere
