U.S. BankCincinnati, OH+3 more
Posted Sep 10
AI Red Team Jobs
AI red teamers probe models and AI products for weaknesses before attackers or ordinary users find them - running adversarial prompts, jailbreak attempts, and structured attacks to surface unsafe outputs, security holes, and policy violations ahead of release.
AI Red Team roles are rare: 3 live right now. Get the new ones every Monday.
Get new AI Red Team jobs in your inbox
Join 100+ AI professionals · Weekly, free, unsubscribe anytime
Latest AI Red Team roles
Why red teaming became a staffed function
AI red teaming stopped being a pre-launch exercise and became a permanently staffed function once system cards and model release documentation turned findings into a published artifact. When a developer commits to describing what adversarial testing was done before a model ships, someone has to run that testing on a schedule, write it up to a standard that survives external scrutiny, and repeat it for every subsequent release. That is a team, not a contractor engagement.
The work splits in two, and the halves hire from different places. Domain-expert red teaming needs people who know a subject well enough to judge whether a model's output is genuinely dangerous or merely alarming, across areas such as chemistry, biosecurity, medicine, law, cybersecurity, child safety, or a specific language and culture, and it does not require machine learning skills. Automated red teaming is ML engineering: generating attacks at scale, training attacker models, and building pipelines that regression-test known failures on every candidate checkpoint.
Agents widened the surface from what a model says to what it does. Testing a chat model means probing for harmful text. Testing a system with tools, browsing, code execution, and persistent memory means probing whether an attacker can make it take an action, such as exfiltrating data, spending money, or modifying state, through content the model merely reads. Hiring in this area is bursty, clustering around release cycles and new product surfaces rather than arriving at a steady rate.
Explore related searches
- Browse the Model Evaluation hub for the full specialization.
- Prefer remote? See remote AI jobs.
- AI Security jobs
- AI Safety jobs
Frequently asked questions
- What does AI Red Team hiring look like at the moment?
- We are tracking 3 live AI Red Team roles across the AI companies we monitor, updated hourly. Each listing links straight to the employer's own application page.
- Do AI red team roles require a machine learning background?
- Not always. Automated red teaming is ML engineering and expects it. Domain-expert red teaming hires for deep subject knowledge, in areas such as security, biosecurity, chemistry, medicine, law, or a specific language and region, and teaches the model interaction on the job. Many teams deliberately mix both profiles, because each finds failures the other misses.
- How is AI red teaming different from a traditional security red team?
- A security red team attacks infrastructure, identity, and applications to reach a defined objective. An AI red team attacks model behavior and policy boundaries, probing for unsafe outputs, jailbreaks, and manipulation of what an agent does. The overlap grows once models are given tools and permissions, and some teams now staff both skill sets side by side.
- What does an AI red teamer actually produce?
- Structured findings rather than a single report: reproducible attack cases with severity judgments, coverage notes on what was and was not tested, regression suites so that a fixed failure stays fixed, and written summaries that feed release decisions and public system cards. Evidence a reviewer can act on matters more than raw volume of jailbreaks.