About NEUROSIS Research
NEUROSIS Research is a small independent AI security research organization. We conduct serious, reproducible work without presenting ourselves as a university, large company, or physical laboratory.
Our legitimacy comes from the work: useful questions, careful experiments, public methods, reproducible results, open tools, and honest engagement with prior research.
Where we started
NEUROSIS began as an experiment in persistent external memory for autonomous agents. It asked what changes when agents stop operating as isolated, ephemeral processes and can retrieve, modify, share, and inherit knowledge over time.
That work led to questions about provenance, trust, memory contamination, Byzantine knowledge propagation, correlated evidence, agent interaction graphs, and adversarial agent ecologies. The concern was always broader than building better memory: what new failures emerge when intelligent systems acquire persistence, shared state, autonomy, and influence over one another?
Why the scope expanded
Modern systems can call tools, change files and infrastructure, use browsers and computers, coordinate with other agents, perceive images and video through VLMs, and control robots through Vision-Language-Action policies. As AI systems acquire more agency, failures cross from generated text into digital and physical action.
NEUROSIS Research follows that progression from memory to agents, multi-agent systems, perception, embodiment, and physical action. The original work remains an active layer of the broader research program.
How we work
We distinguish hypotheses, experimental observations, confirmed vulnerabilities, negative results, and future risks. Failed hypotheses are part of research. Claims that cannot be reproduced are not treated as findings.
Research notes, experiments, tools, datasets, benchmarks, and papers are made public when practical and compatible with responsible disclosure. Each published artifact should state its maturity, method, evidence, limitations, and reproduction path.