LLM Agent Safety Research
Aether is an independent LLM agent safety research group dedicated to conducting impactful research that ensures the responsible development and deployment of AI technologies. We work on whatever seems most impactful to us, focusing on critical areas that can positively influence AGI companies, governments, and the broader AI safety field.
Our primary research focus has been on chain-of-thought monitoring. We investigate how information access affects LLM monitors' ability to detect sabotage and other safety-critical behaviors. We've also developed a taxonomy for understanding hidden reasoning processes within LLMs, providing a structured framework for analyzing covert reasoning mechanisms.
We're exploring topics including shaping the generalization of LLM personas, interpretable continual learning, and pretraining data filtering. Our research agenda remains flexible to focus on the most impactful projects.
Applications reviewed on a rolling basis • Apply early
So far, we have focused on chain-of-thought monitoring. See our Research section for details on our work, including our forthcoming paper How does information access affect LLM monitors' ability to detect sabotage? and our post Hidden Reasoning in LLMs: A Taxonomy.
We are not committed to a specific research agenda for the upcoming year yet. Topics we're currently exploring include shaping the generalization of LLM personas, interpretable continual learning, and pretraining data filtering. We plan to always work on whatever seems most impactful to us.
We generally prefer that candidates join us for a short-term collaboration (1-3 months part-time) to establish mutual fit before transitioning to a long-term position. However, if you have AI safety experience equivalent to having completed the MATS extension, we are happy to interview you for a long-term position directly. The interview process will involve at least two interviews: a coding interview and a conceptual interview where we'll discuss your research interests. The expected starting date for long-term researchers is Feb-May; we're happy to start short-term collaborations as soon as possible.
If you are only interested in short-term collaborations, you can fill out this form instead.
Founder & Researcher
Researcher
Researcher
Astera Institute
Apollo Research
Google DeepMind
Independent / LawZero