Vérifiez votre IA fournisseur, modèle, conformité, risques, sous-traitants, transparencefournisseurmodèleconformitérisquessous-traitantstransparencefournisseur

TOPdesk is service management software from TOPdesk in Delft, used by IT, facility and other service desks to handle tickets, changes, assets and a self-service portal. Universities, colleges and schools use it as the service desk for students and staff, for example for password resets, ID cards and onboarding. The vendor says its built-in AI suggests ticket categories, priorities and operator groups, summarises tickets, drafts replies and knowledge items from the knowledge base, improves and translates text, and powers search in the portal, and that its AI teammate TOPdesk Robin answers end users' questions in the portal and helps operators inside tickets. AI features are off by default, the operator decides on triage suggestions and reviews anything sent to customers, and AI output is labelled.

Writefull is an AI writing assistant for academic English, part of Digital Science. Students and researchers use it in Microsoft Word, in Overleaf and on the web for language feedback, paraphrasing, title and abstract generation and a language check before submission; institutions buy licences for staff and students, and publishers use it in copyediting. The vendor says its own language models, trained on published research papers, give the English language edits and run the Title and Abstract Generators, Academizer, Paraphraser and a GPT Detector, while features such as paraphrase, summarise, explain, Ask AI and TeXGPT send text to OpenAI (GPT) or in some cases Anthropic (Claude) after the user agrees. Writefull says it does not store users' texts or train on them, that OpenAI and Anthropic delete text within 30 days and do not train on it, and that institutions can switch individual AI features off. It warns that its language models can make mistakes and that users should review every suggestion before accepting it.

Dimensions is Digital Science's research database and analytics platform, linking publications, grants, clinical trials and patents. Universities, research funders and research offices use it for literature discovery, funding and portfolio analysis and research evaluation. The vendor says its AI features include one-click AI summaries of records in search results (in the free and paid versions), Natural Language to Query, which turns plain-language questions into searches in Dimensions Analytics, and Research GPT, a ChatGPT-based assistant built with OpenAI that answers from Dimensions content with links to the sources; the Enterprise version also covers grants, clinical trials and patents for subscribers with ChatGPT Enterprise. Digital Science says these tools lower but do not remove the risk of hallucinations and give click-through references so that users can verify answers.

Imagetwin is an online service and API from ImageTwin AI GmbH in Vienna that screens images in research papers for integrity problems. Publishers, universities, research offices and individual researchers use it to check manuscripts, theses and student submissions before submission or publication. The vendor says its AI extracts and compares all sub-images to find duplicates even after rotation, scaling or colour changes, checks figures against more than 160 million published scientific images, detects splicing and other manipulations, and detects AI-generated images with a confidence score and attribution to the generating model. Results are highlighted for a reviewer to verify, and organisations can build private image databases to compare new submissions against.

Proofig AI is an online service from Proofiger Ltd. in Israel that checks the images in scientific manuscripts for integrity problems before or after publication. Publishers, journals, universities and research institutes use it to screen life-science figures such as microscopy, Western blots, gels, FACS and histology images. The vendor says its proprietary image processing and AI detect duplication, manipulation, image plagiarism against published material (including the PubMed open-access database) and AI-generated images, and produce a report of suspected issues. Reviewers check each flagged item with forensic tools and decide what to include; the vendor says such tools should surface suspected issues for human review rather than deliver verdicts.

Anthropic's small, fast Claude 4.5-generation model. Named as the model behind AI features in Canvas (Instructure IgniteAI Nutrition Facts) and Turnitin Clarity / Feedback Studio (Turnitin administrator FAQ and trust center).