AI Use on NCShare
NCShare is a shared research and teaching environment funded by the National Science Foundation and operated by a consortium serving many North Carolina institutions. Use of AI tools on NCShare must be consistent with research and instructional purposes, must respect data, privacy, and access requirements, and must not interfere with the availability, integrity, or security of NCShare systems or other users' work.
NCShare accounts are provided to authorized users at participating institutions. Your use of NCShare remains subject to the computing and acceptable use policies of your home institution, in addition to the expectations on this page. If the two ever conflict, follow the stricter one and contact your local support team.
Supervised use
AI-assisted development and research workflows are appropriate on NCShare when they are run transparently, within your allocation, and with normal user oversight. Avoid running autonomous or agentic AI tools in ways that grant broad, unattended control over files, shell commands, network access, credentials, job submission, or other NCShare resources.
Agent commands must run on compute nodes rather than the login node. See Access via AI Agents for the SSH host alias that arranges this.
Prohibited and discouraged tools
OpenClaw is not permitted on NCShare. Users must not install, run, expose, proxy, or connect OpenClaw services from the login node, compute nodes, interactive sessions, Open OnDemand environments, or NCShare storage.
Users are strongly discouraged from running AI tools in unsafe permission-bypass modes, including flags or settings such as --dangerously-skip-permissions, bypassPermissions, YOLO mode, blanket tool approval, unrestricted shell execution, or disabled approval prompts. Such modes are especially risky on shared HPC because an agent acts with the same access as your account, including your files, credentials, job queues, network access, and any shared research data you can reach. Anthropic's Claude Code permission modes documentation describes bypassPermissions as disabling permission prompts and safety checks so tool calls execute immediately, and states that --dangerously-skip-permissions is equivalent.
Data
No sensitive data is permitted on NCShare, and that applies equally to data an AI agent reads, writes, or transmits. Do not point an agent at data you could not place on NCShare yourself, and do not configure agents to send NCShare data to external services beyond what your workflow requires and your institution permits.
NCShare storage is not backed up and is not intended for long-term use. Files on /work older than 75 days are purged automatically. An agent that writes results only to cluster storage has not preserved them.
Impact on shared resources
AI tools must not degrade the performance, stability, availability, or security of NCShare for other users. Jobs, processes, agents, services, or accounts that overload systems or networks, consume excessive resources, create security concerns, or otherwise disrupt other users' work may be limited, suspended, or terminated. If support staff raise unsafe, disruptive, or noncompliant AI use with you and the behavior continues, your access may be suspended pending review by your institution's representative.
Responsibility and recovery
Poorly supervised AI use can create serious operational and data risks. If an agent runs in an unsafe mode, with overly broad permissions, or in a way that allows destructive or large-scale changes, NCShare and institutional IT staff may not be able to help you fully recover. This includes loss or corruption of files, deletion or modification of code or research data, unintended job submissions, credential exposure, and changes to software environments.
You remain responsible for activity initiated from or attributable to your NCShare account. Labs and courses are responsible for maintaining their own copies of important work, supervising AI-assisted workflows, and making sure members understand these risks before using AI coding agents on shared infrastructure.
The practical expectation
Use AI tools on NCShare only in ways you can supervise, explain, and defend. Do not delegate unrestricted control of shared computing resources to an agent. Keep approval prompts on, keep a human in the loop, run heavy work through Slurm, protect your credentials, keep copies of anything you cannot lose, and talk to your local support team before attempting higher-risk AI workflows.