Terms of Use KI-Toolbox
Scope and Purpose
The KI-Toolbox is a service provided by the Scientific Computing Center (SCC) at the Karlsruhe Institute of Technology (KIT) for the use of various generative AI (artificial intelligence) models in the context of work-related or academic activities. The service is intended for KIT employees, KIT students (starting in April 2026), and users with a Guest and Partner (GuP) account.
The provisions of KIT’s Information and Communication Technology (IuK) Regulations regarding permissible use, user obligations, operator authority, measures in case of violations, logging, and liability apply. These Terms of Use supplement the KIT ICT Regulations exclusively with AI-specific provisions (data categories/risk classes, choice of model—local vs. Azure, use for examinations, export controls, and special data protection considerations).
Service Description, Models, and Providers
The KI-Toolbox provides both open-source models operated at the SCC (“local models”) and commercial models hosted by cloud providers (“external models”). The models are labeled accordingly in the KI-Toolbox interface.
Local models vary depending on availability and version. The decommissioning of local models will be announced in advance and communicated via SCC notification channels. Additional information channels include the Matrix channel for the KI-Toolbox and the “ki-toolbox-news” channel integrated into the KI-Toolbox.
The external models from OpenAI (“GPT”) are hosted on the Microsoft Azure Cloud (hereinafter “Azure”). The external models from Google (“Gemini”) and Anthropic (“Claude”) are hosted on the Google Cloud Platform (hereinafter “Google”). Requests are processed in Azure or Google data centers exclusively within the EU. The use of content for training purposes is permanently disabled.
The service is operated on a “best effort” basis—high availability and performance are targeted, but without an SLA or availability guarantee.
Access, Authentication, and Accounts
Access to the KI-Toolbox is granted via the KIT account using single sign-on (OpenID Connect). Two-factor authentication is not currently required.
Currently, the only roles supported in the KI-Toolbox are standard users and administrators. Groups for the KI-Toolbox can be created via group management; group memberships are synchronized upon login. Groups with special permissions are technically possible and can be assigned by the SCC in special cases.
Data Categories, Protection Classes, and Model Selection
Personal data (Art. 4(1) GDPR) may be processed exclusively using models hosted locally at KIT. Processing of personal data using external models via Azure or Google is prohibited.
The following is decisive with regard to KIT’s risk classification: https://www.isb.kit.edu/134.php.
- High-risk use: exclusively with locally hosted models
- Very high-risk use: prohibited
Users are responsible for ensuring lawful processing, selecting an appropriate model, and implementing the necessary protective measures.
It is prohibited to enter information subject to export controls, sanctions, or other security restrictions into the KI-Toolbox.
Logging, Storage, Deletion, and Backups
Chat histories, uploaded files, and AI-generated outputs are stored in the KI-Toolbox on a user-specific basis for one year; users can delete these on their own at any time. Upon deactivation of the account (e.g., upon leaving KIT), chat histories are deleted within 24 hours at the latest.
For capacity and billing purposes, the metadata of a request (user ID, time, selected model, number of input/output tokens) is stored on SCC servers when a request is submitted. This metadata is not combined with other data. This data is retained for one year. The content of the requests and the responses are not stored. Users cannot delete this metadata themselves.
System backups with standard rotation and retention periods may result in deleted data persisting in backups for a limited time. Operational access occurs only for recovery and security purposes.
Training, Fine-Tuning, and Evaluation
All user content (prompts, uploads, outputs) is treated as strictly confidential by the SCC and is not used for training, fine-tuning, evaluation, or research purposes.
For models provided via Azure and Google, training use by third-party providers is disabled.
Copyright and Ownership of Results
Users are responsible for complying with copyright, license, trademark, personality rights, and trade secret protection laws, as well as other third-party rights. The relevant KIT regulations apply.
Generated content may be inaccurate, incomplete, or legally problematic. Users must conduct an appropriate technical and legal review before further use.
Users retain rights to their own input. Users acquire rights of use to generated content to the extent permitted by law; third-party rights remain unaffected. License terms of individual open-source models must also be observed, if applicable.
Permitted Use
The KI-Toolbox may be used exclusively for work-related or academic purposes. Use for private purposes is not permitted.
Use in examination contexts is permitted only with the express permission of the responsible examination officials and within the framework of the applicable examination regulations.
The creation and use of manipulative AI content (e.g., deepfakes) without proper labeling or authorization is prohibited. The selection of the model must be made strictly in accordance with the section “Data Categories/Risk Classes.”
Costs and Budgets
Costs are incurred when using external models within the KI-Toolbox. All users have a monthly “standard budget” of 5 USD per month (effective July 1, 2026). Costs incurred within the standard budget are currently covered centrally and are not billed individually. If the budget is exceeded, a corresponding error message will be displayed when using a model (“ExceededBudget: End User=xy1234∂kit.edu over budget. Spend=xxx.xxxxxxx, Budget=5.0”). The budget is reset for all users at midnight (UTC) at the beginning of each month. Current usage can be viewed at any time in MyLiteLLM.
For needs that exceed the standard budget, KIT employees can adjust their individual budgets via MyLiteLLM. Before changing the budget, approval must be obtained from the budget managers or from the executive level! The SCC bills individual budgets monthly via the user’s cost center, charging the full amount of the actual costs incurred. Students and users with guest and partner accounts cannot set individual budgets.
Use of local models is not charged to the budget or is billed at 0 USD. If the budget is exceeded, use of local models is still possible.
Rate Limits and Fair Use
There is currently no rate limit. All users are required to use resources fairly to ensure availability for everyone. The SCC reserves the right to set individual rate limits in cases of excessive usage by individual users.
User Responsibility
There is no substantive or legal review of inputs or outputs. Generated content may be incorrect or misleading and must not be used without careful consideration. Users are responsible for verifying the accuracy of the content and for its further use.
data protection
The privacy policy for the KI-Toolbox applies.
Governance, Compliance, and Documentation
The provisions of KIT’s ICT Regulations apply. Furthermore, KIT’s AI guidelines (https://www.kit.edu/downloads/KI-Guidelines-de.pdf) and KIT’s Risk Classification (https://www.isb.kit.edu/134.php) must be observed.
There is no automatic archiving of work-related results. Users must ensure that any required archiving and documentation is carried out (e.g., exporting chat histories).
Updated on July 1, 2026