Member of Technical Staff, Automating AI Safety Research
Trajectory Labs, Berkeley
Trajectory Labs, Berkeley
Trajectory Labs is hiring researchers to build the evals and training data to measure and improve the ability of AI models to perform AI safety work. These are rubric-based evaluations of fuzzy tasks that use your AI safety expertise.
You'll lead R&D on new eval products, build the pipelines to produce them at scale, and evaluate the reliability of these evaluations.
Some of our work is directly for the public benefit, so you will likely publish some benchmarks publicly. In other cases, you’ll lead a team to deliver a product to the frontier labs who will use it. You will also have opportunities to identify new product lines that will advance the company’s goals, and explore them.
Frontier models already solve coding problems that take humans days; within years, they may run AI R&D autonomously. Over time, we want AI labour to be used effectively towards AI safety work, and this requires AI models to get better at some relevant vague tasks https://arxiv.org/abs/2605.06390, like performing interpretability research, reasoning about safety cases or reasoning about risks from future powerful systems. We're partnering with frontier AI labs to close that gap.
We already build conceptual reasoning evals for frontier labs, written by expert researchers in alignment, governance, and macrostrategy, and we're now growing the team that builds them. We're hiring full-time researchers to take these methods further: to new kinds of task, such as building evaluations to improve interpretability methods, evaluating research tasks that have never been done before, anti-slop evals to improve our ability to understand and oversee models, and performing alignment auditing.
Each new line of work needs someone to work out how to measure it, show that the measurement holds up, and get it to the people who will use it, whether by publishing it or by building it into a product for labs.
It's important to us that our work is robustly positive, so we focus our evals on the capabilities that differentially accelerate AI safety, not capabilities in general. While we take seriously the risk of spillover into other capabilities, we think it is crucial that AI models can help with safety and alignment in time, and that this outweighs the second-order effects.
You will also do some of the following, but which ones will depend on your skill set:
We're hiring AI alignment and AI evaluations researchers who want to measure and improve the AI safety capabilities of frontier models. Beyond building evals, you'll either write research up to an academic standard or lead a team that takes a product to customers; we expect most strong candidates to be better at one than the other.
You'll work on the methods that decide whether AI safety research can be measured well enough for labs to automate it, alongside a team of expert researchers who write and review our evals. The team is small and growing, so you could expect to lead a product line of your own.
Your colleagues include the lead author of the METR time horizons paper, the former head of AI safety at Oxford's Future of Humanity Institute, and many other PhDs and research fellows from Oxford University who are leaders in their fields.
Total compensation: Includes base salary, bonus, and the estimated annualized value of equity. Your level depends on the scope, autonomy, and impact we expect you to have:
Base salary: $200,000–$400,000 per year.
Location and workspace: Our team works out of Constellation in Berkeley, CA, and we prefer someone who can work alongside us there. We are open to remote for the right candidate.
Reports to: Ben West, CTO.
Our mission is to automate AI safety https://www.lesswrong.com/posts/F3j4xqpxjxgQD3xXh/ai-for-ai-safety, to pave the way for a future where the vast majority of AI safety work is done by AI models.
To that end, we build safety and alignment evals, red-teaming programs, and RL environments for frontier AI labs. Our areas of focus include red teaming, control, anti-distillation, AI safety reasoning, and embedded assessments.
Although Trajectory https://www.trajectorylabs.com is only a year old, we have already been featured in Anthropic's Claude Opus 5.5 System Card https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf#page=59 (the only commercial red team to jailbreak its cyber safeguards) and Google DeepMind's Gemini 4 Argon Safety & Security Report https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf, led Anthropic's prompt injection evaluation of Auto mode https://claude.com/blog/auto-mode-default-in-claude-code, and been featured in Meta's Muse Spark Safety & Preparedness Report https://ai.meta.com/static-resource/muse-spark-safety-and-preparedness-report/.
$380K–$500K / yr
Posted Sep 3, 2026
AnthropicNew York City, NY +2
$320K–$485K / yr
Posted 9 days ago
Thinking Machines LabSan Francisco
$350K–$475K / yr
Posted Aug 24, 2026
Character.AIRedwood City, CA
$225K–$400K / yr
Posted Oct 3, 2025
MetaMenlo Park, California +1
$184K–$257K / yr
Posted 21 days ago
checked 8 days agoTrajectorySan Francisco
Pay not posted
Posted May 26, 2026
Source: the Trajectory Labs Ashby job board. We last checked this job 50 min ago. We first saw it 50 min ago. A job on a job board does not always mean that the team interviews people now.
No one has reported an interview at Trajectory Labs yet. Did you interview there? Add the questions.
This base pay is above the median for AI research jobs in the Bay Area with no level in the job title.
We compare the middle of each posted range. This is base pay as posted. The group does not include this job.