DFKI Salaried PhD in Mechanistic Interpretability for Safe Agentic AI
German Research Center for Artificial Intelligence (DFKI)
DFKI is recruiting a full-time PhD researcher in Saarbruecken for a 36-month salaried project on mechanistic interpretability and fine-grained steering of language models for safe agentic AI. Applications close September 15, 2026.
The German Research Center for Artificial Intelligence (DFKI) is recruiting a full-time Researcher/PhD Student for its Machine Learning and Testing (MLT) department in Saarbruecken. The temporary employment contract is for 36 months and can begin at the earliest possible date.
The project develops fine-grained, interpretable steering methods that control language-model behavior without retraining. Research will examine activation-level interventions, whether safety-relevant and task-critical representations can be separated, and how steering methods scale across models and agentic settings.
The doctoral researcher will work in the Efficient and Explainable NLP team under Dr. Simon Ostermann, within MLT directed by Marius Mosbach, with support from the SAINT and Neuro-mechanistic Modeling groups. Expected outputs include high-quality research submissions to venues such as NeurIPS, ICML, ACL and ICLR.
Employment and funding: DFKI describes the role as full-time, salaried employment with competitive, market-based compensation. The exact salary is not published. The role includes access to GPU infrastructure and support for publications and conference participation.
How to apply: submit the official DFKI online application by September 15, 2026, quoting reference 11387-2026. Questions may be directed to Dr. Simon Ostermann at Simon.Ostermann@dfki.de.
Source and verification
Check the current official call before applying
This independent listing was last checked on September 1, 2026. We compare the named provider, deadline status, eligibility, funding claims and application destination where those details are available. The live provider page is the final authority.
Benefits
- Competitive, market-based salary under a 36-month full-time employment contract; exact amount not published.
- Access to modern GPU clusters and DFKI research infrastructure.
- Support for conference participation and research publication.
- Interdisciplinary supervision across MLT, SAINT and Neuro-mechanistic Modeling.
- Employee benefits may include Urban Sports Club access, corporate benefits and a Jobticket subsidy.
Eligibility summary
- Open to qualified applicants of any nationality; the vacancy does not state a nationality restriction.
- Applicants must satisfy DFKI employment and doctoral-research requirements and be available for full-time work in Saarbruecken, Germany.
- DFKI encourages applications from women in science and gives preference to applicants with disabilities when equally qualified.
Requirements
- Master degree in computer science, natural language processing, machine learning or a closely related field.
- Strong foundation in deep learning and language models.
- Experience with mechanistic interpretability methods such as circuit analysis, sparse autoencoders or activation patching.
- Strong Python and PyTorch skills.
- Ability to conduct independent research and collaborate across research groups.
- Experience with model steering, causal intervention, TransformerLens or SAE libraries, relevant publications, or AI-safety research is desirable.
Documents required
- Completed DFKI online application form.
- Curriculum vitae.
- Cover letter or statement of motivation.
- Degree certificates and academic transcripts.
- Publication list or research outputs where applicable.
- Any additional documents requested by the DFKI application portal.
Language/test requirements
How to submit applications
Provider: German Research Center for Artificial Intelligence (DFKI)
Official DFKI vacancy reference 11387-2026. Full-time, temporary, 36 months. Deadline September 15, 2026.
Listing-board disclaimer
Prime Scholarship Alerts is an independent listing board, not the scholarship provider, university or funder. We do not select applicants, collect provider application fees or guarantee an award. Verify requirements, deadlines, funding, fees and application instructions through the current provider page before applying.