NEUROSIS Research

Description

NEUROSIS Research is a small, independent organization dedicated to AI security research. It focuses on vision-language models, Vision-Language-Action systems, robots, autonomous agents, and machine memory. The organization began as an experiment in persistent external memory for autonomous agents, exploring how agents accumulate, retrieve, modify, and share knowledge over time. This research led to broader questions about provenance, trust, memory contamination, and adversarial agent ecologies.

The research program at NEUROSIS is centered around the security of intelligent systems whose perceptions can translate into physical actions. Key areas of study include adversarial visual inputs, physical prompt injection, scene manipulation, and grounding failures in vision-language models. The organization also investigates action freshness, physical TOCTOU (Time-Of-Check to Time-Of-Use) attacks, and runtime validation in robotics.

NEUROSIS Research emphasizes empirical and reproducible research, making its findings public when responsible disclosure allows. The organization distinguishes between hypotheses, experimental observations, confirmed vulnerabilities, and negative results. Research notes, experiments, tools, datasets, and papers are shared to promote transparency and collaboration.

The NEUROSIS Research public memory stores anonymous, immutable plaintext records, accessible via a public API without requiring an account or API key. This API allows users to create and read records, with strict guidelines to ensure data integrity and security. The organization is committed to studying the complete chain of memory failures and embodied failures in persistent embodied agents and robot fleets.

NEUROSIS Research's Core Features

• Independent AI security research

• Focus on vision-language models

• Study of Vision-Language-Action systems

• Research on autonomous agents

• Investigation of machine memory

• Public memory API available

• Empirical and reproducible research

• Open and evidence-aware publication

How to use NEUROSIS Research?

1. 1

   Explore: Visit the NEUROSIS Research website

2. 2

   Read: Access research papers and findings

3. 3

   Engage: Participate in public discussions

4. 4

   Contribute: Use the public memory API

NEUROSIS Research's Use Cases

• AI Security Research
• Vision-Language Models
• Robot Security
• Autonomous Agents
• Machine Memory

FAQ from NEUROSIS Research

• What is NEUROSIS Research?

  NEUROSIS Research is an independent organization focusing on AI security research, particularly in vision-language models and autonomous systems.

• What are the main research areas of NEUROSIS?

  The main research areas include VLM security, VLA and robot security, agent and multi-agent security, and memory and provenance.

• How does NEUROSIS Research share its findings?

  NEUROSIS Research shares its findings through public notes, experiments, tools, datasets, and papers, ensuring transparency and reproducibility.

• What is the NEUROSIS public memory API?

  The public memory API allows users to store and retrieve anonymous, immutable plaintext records without requiring an account or API key.

• What is the focus of the StaleAction project?

  StaleAction focuses on physical TOCTOU attacks against Vision-Language-Action robots, studying how perception and action can be compromised.

• How does NEUROSIS Research ensure responsible disclosure?

  NEUROSIS Research makes work public when responsible disclosure and safety permit, clearly labeling hypotheses, observations, and findings.

• What is the research philosophy of NEUROSIS?

  The research philosophy emphasizes empirical, reproducible, adversarial, systems-oriented, and open and evidence-aware approaches.

• What are the operational limits of the NEUROSIS API?

  The API has configurable operational limits, including JSON body size, content size, unique references, and per-minute request limits.