Independent AI safety research
Understand
what AI
actually does.
Investigating AI behavior, misuse, and emerging risks—through evidence people can examine.
Explore our publications
Evidence for researchers.
Clarity for policymakers.
01 / For researchers
Technical Research
Methods, investigations, and analysis. Follow the evidence behind our understanding of AI.
Explore technical research →Publications will appear here as they are released.
02 / For policymakers
Policy Briefs
Key findings and policy implications. Understand what the research means for decisions about AI.
Read policy briefs →Publications will appear here as they are released.
01 / Research agenda
A closer look at AI.
Understanding what systems do, how they are used, and where oversight falls short.
01 / AI behavior & oversight
Beyond the benchmark.
What happens when an AI system leaves a controlled test and takes on a longer, messier task? We study the behavior that evaluations can miss.
Explore this research direction
Questions in this research agenda:
- Which behaviors change as tasks become longer and more open-ended?
- What can monitoring reveal—and what does it miss?
- How can evaluations better reflect real conditions?
02 / AI misuse & investigations
Follow the evidence.
How are AI systems being used in the wild? We connect observable actions, infrastructure, and artifacts to investigate misuse and emerging risks.
Explore this research direction
Questions in this research agenda:
- Where does AI meaningfully change harmful activity?
- What evidence connects a system to an observed action?
- Which signals justify closer investigation?
02 / Our approach
Start with the record.
A signal is a starting point. Understanding comes from reconstructing what happened, testing explanations, and making the limits visible.
01 / Observe
Reconstruct
what happened.
Begin with a record of actions: the sequence, the tools used, and the artifacts left behind. Preserve the context that makes each observation meaningful.
What do we actually know?
03 / Why Noetic
AI is changing.
Understanding
has to keep up.
Noetic Research brings an investigative lens to AI safety.
As AI systems become more capable and more widely used, their behavior becomes harder to understand from a benchmark alone. The important questions extend beyond what a model can do to what it actually does—and in whose hands.
Our focus is the evidence between those questions: observable behavior, patterns of use, and the limits of oversight. We aim to turn that evidence into research that is useful to the people building, studying, and governing AI.
Explore our research agendaHow we work
Research others can examine.
Behavior before inference.
Distinguish what a system did from what we think it intended. Treat attribution and explanations as claims to test.
Make the reasoning visible.
Show the path from observations to conclusions, with methods and context that let others evaluate the work.
Keep uncertainty explicit.
Say what the evidence supports, what it leaves open, and what would change our assessment.