About the role
Safeguards builds the systems that keep misuse off the platform. That means classifiers on inputs and outputs, usage monitoring, account-level enforcement and the review tooling that lets human investigators act on what the models find.
As Frontier AI ships more capable models, the safeguards must scale with them, and they must work for agents as well as chat. Clexerl Go runs autonomous actions on behalf of users, so we need to detect harmful intent across long tool-using sessions and not only single prompts. Precision matters as much as recall: false positives block legitimate work and erode trust.
We iterate against the red team's findings and measured abuse patterns. Every detector has an owner, an evaluation set and a defined path for appeals.
What you’ll do
- Train and deploy classifiers for harmful content and misuse, with measured precision and recall on held-out and adversarial sets.
- Build real-time and offline monitoring pipelines that detect abuse patterns across accounts, sessions and agent tool use.
- Design enforcement systems, including graduated responses, rate limits and account actions, with audit trails and an appeals path.
- Work with the red team to turn discovered jailbreaks and capability findings into detectors and mitigations.
- Build investigator tooling so analysts can triage flagged activity quickly and feed labels back into training.
- Track drift and adversarial adaptation, and retrain or retune before detection quality degrades.
What you’ll bring
- Experience building and operating machine learning classifiers in production, ideally in trust and safety, fraud or abuse detection.
- Strong Python and data engineering skills, including streaming pipelines and large-scale log analysis.
- Understanding of adversarial behaviour: evasion, prompt injection and attempts to probe detector boundaries.
- Rigour in evaluation, including labelling quality, threshold selection and cost-sensitive trade-offs.
- Care for privacy and proportionality when working with user data.
- Ability to work with legal, policy and operations colleagues on decisions that are not purely technical.
Nice to have
- Experience with language-model-based classifiers, constitutional or rule-driven grading, and distillation into cheap runtime detectors.
- Background in abuse investigation or threat intelligence.
- Familiarity with detecting coordinated inauthentic behaviour across accounts.
- Experience building human-in-the-loop labelling and review systems.
Work you might take on
- 01
Build a classifier cascade where a cheap model screens all traffic and a larger model reviews only the ambiguous cases.
- 02
Detect multi-turn misuse in agent sessions by scoring the sequence of tool calls, not just individual requests.
- 03
Create an adversarial evaluation set from red team jailbreaks and track detector recall on it for every release.
- 04
Design the appeals workflow and audit log so a wrongly actioned account can be reviewed and restored quickly.
Compensation
$250K – $400K
The US base salary range for this role. Every offer also includes equity in Clexerl and, when it spins out, in the venture you help build. We tell you the reasoning behind your number.
Benefits
Health, fully covered
Medical, dental and vision for you and your dependants, premiums paid in full.
Two kinds of equity
Equity in Clexerl, plus a stake in the venture you help build when it spins out.
22 weeks parental leave
Fully paid, for every parent, however your family arrives.
Rest that sticks
Flexible time off with a four-week minimum we actually check on, plus a company-wide week in December.
Learning and compute
A yearly budget for courses, conferences and books, and compute credits for side projects.
401(k) with a 4% match
Matched from your first paycheque and vested immediately.
How we hire
Apply 5 minutes
Send links to what you have built and a few lines on why this role. No cover letter needed.
Intro call 30 minutes
A conversation with the hiring manager about the work, the team and what you want next.
Work sample 2–4 hours
A paid take-home or a pairing session on a real-shaped problem. You pick the format.
Studio day One day
Meet the team, go deep on your work and ours. Remote or in person.
Offer Within 5 days
A written offer with the full range, the equity and the reasoning behind both.
Opens an email with a short template. Links beat résumés.