Aether is an independent research lab focused on LLM agent safety, conducting technical research on the alignment, control, and evaluation of large language model agents.
Aether is an independent research lab focused on LLM agent safety, conducting technical research on the alignment, control, and evaluation of large language model agents.
People
Updated 05/18/26Funding Details
Updated 05/18/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- -
Org Details
Updated 05/18/26Aether is an independent research lab founded in 2024 by Rohan Subramani, focused on the safety of large language model (LLM) agents. The organization's mission is to conduct technical research that yields valuable insights into the risks and opportunities that LLM agents present for AI safety, with the ultimate goal of ensuring the responsible development and deployment of AI technologies. Aether began accepting applications in August 2024 for a founding cohort of 4-6 full-time researchers, initially targeting in-person work in Berkeley or London. By mid-2025, the core team had settled on London, and by late 2025, had relocated to Trajectory Labs, a nonprofit AI safety coworking space in downtown Toronto, Canada. The team as of early 2026 consists of three full-time researchers: Rohan Subramani (founder), Rauno Arike, and Shubhorup Biswas. Aether's primary research focus is chain-of-thought monitorability — investigating how information access affects LLM monitors' ability to detect sabotage and other safety-critical behaviors. The group has also developed a taxonomy of hidden reasoning mechanisms in LLMs and is exploring adjacent directions including interpretable continual learning, shaping LLM persona generalization, and pretraining data filtering. The organization is advised by Seth Herd (Astera Institute), Marius Hobbhahn (Apollo Research), Erik Jenner (Google DeepMind), Francis Rhys Ward (LawZero), and Zhijing Jin (University of Toronto). Rohan Subramani began a PhD at the University of Toronto in September 2025 under Professor Zhijing Jin, continuing his Aether work within his PhD program. Aether has published research on LessWrong and the EA Forum, maintains an active Discord community, and has conducted multiple hiring rounds to grow its research team.
Theory of Change
Updated 05/18/26Aether believes that LLM agents represent a near-term, high-stakes safety challenge that is underserved relative to its importance. By producing rigorous technical research on how to monitor, align, and control LLM agents — particularly by improving chain-of-thought monitorability and understanding hidden reasoning — Aether aims to generate insights that can be adopted by frontier AI labs, governments, and the broader AI safety field. The goal is to make advanced AI systems more legible and controllable, reducing the risk that autonomous AI agents pursue goals misaligned with human values.
Grants Received– no grants recorded
Updated 05/18/26Projects– no linked projects
Updated 05/18/26Discussion
No comments yet. Be the first to share your thoughts.