The ISSAI AI Computing Cluster consists of NVIDIA DGX systems operated as standalone machines. Each machine is a shared resource: its GPUs, CPU cores, RAM and storage are used by several users at the same time, so the rules below must be followed to avoid interfering with the work of other users. Please be as accurate as possible about the resources which your jobs will require, as this will help the system run more efficiently and may help your jobs run more quickly.Failure to do so could result in poor performance and may adversely impact other users’ work.

USER RESPONSIBILITY FOR DATA AND DATA RETENTION POLICY

• Each user is provided with a personal home directory, /home/<user name>, with limited capacity, intended for personal files, source code, and configuration data, and a significantly larger working directory, /raid/<user name>, intended for datasets, job results, and interim data generated during job execution.
• The standard storage quota on /raid/<user name> is up to 2 TB per user. The quota may be increased upon a separate agreement with ISSAI, subject to available storage capacity.
• Users must regularly remove unnecessary files, temporary data, intermediate results, logs, and other files that are no longer required.
• Users are solely responsible for their own data, including its integrity, availability, confidentiality, backup, and recovery. ISSAI does not guarantee the recovery of user data in the event of hardware failure, software failure, accidental deletion, corruption, or other technical incidents.
• The DGX system storage should not be considered a permanent or archival storage service. Users must maintain their own backup copies of important data on external or institutional storage.
• ISSAI may perform system-level backups of cluster infrastructure and/or selected data where such backup services are available. However, such backups do not replace the user’s responsibility to maintain independent copies of important data.
• Backup frequency, retention period, and the scope of data included in system backups may vary depending on the storage system and operational requirements. ISSAI does not guarantee that every file stored by a user will be included in backups.
• When a user’s account is deactivated — for example, when the user leaves the university or the research group, or requests account closure — the data in /home/<user name> and /raid/<user name> is retained for 6 months from the date of deactivation. After this retention period, both directories are permanently deleted to free resources for active users.
• Before deletion, ISSAI sends a reminder to the user’s e-mail address on record where reasonably possible. Users are responsible for keeping their contact e-mail up to date and for copying all data they want to keep to another storage location before the retention period expires. After deletion, the data cannot be recovered.
• Users who plan to leave the university or to stop using the cluster are encouraged to transfer the data they want to keep — for example, to the ISSAI S3 Object Storage or to external or institutional storage — before their account is deactivated.
• Data stored on the DGX systems that is required for ongoing research must be actively maintained by the user. ISSAI reserves the right to remove data that violates storage policies, exceeds the assigned quota, or significantly affects DGX operation.
• In exceptional circumstances, including storage capacity constraints, security incidents, hardware failures, or maintenance requirements, ISSAI may temporarily restrict access to or remove data in order to maintain the stability and security of the DGX system.
• Users must ensure that they have the necessary rights, permissions, and licenses to store and process their data on the DGX system.
• Users must not store sensitive, confidential, or personal data unless such storage and processing has been explicitly authorized by ISSAI and appropriate security requirements have been agreed upon.

AI AGENTS POLICY

• AI agents, including autonomous coding assistants, workflow automation agents, and similar agentic AI tools, are powerful systems that can introduce security risks and cause disruptions for the user and for other users.
• AI agents may execute commands without direct supervision, access data beyond their intended scope, consume shared resources (GPU memory, CPU cores, storage) unpredictably, and interfere with the work of other users.
• AI agents may be used on DGX machines and HPC cluster login nodes subject to the following restrictions: they must operate in interactive mode under the user’s direct supervision. Unattended or fully autonomous operation is prohibited. Users are encouraged to develop and test agentic workflows on their own computers before deploying them on the DGX systems.
• Users are not allowed to run coding agents with options that disable confirmation prompts or bypass safety checks, such as “never ask”, auto-approve, auto-accept, or “skip dangerous commands” modes (for example, the –dangerously-skip-permissions or –yolo flags).
• All potentially destructive or security-relevant actions performed by a coding agent must remain subject to explicit confirmation by the user.