AI Challenge Competition

The AI-BOOST AI Challenge Competition is an EU‑funded two‑phase open innovation prize competition (Spark Phase and five‑month Advance Phase) offering Spark awards of €28,500 and a final prize of €100,000 plus special awards, with a live demonstration in Brussels in February 2027. Two highlighted challenges seek Generative AI solutions: Challenge 3 (EUCAIM) for generating realistic synthetic clinical imaging cohorts to improve dataset completeness, balance and harmonisation, and Challenge 4 (Siemens) for automated generation of simulation‑ready driving scenarios from crash data and standards. Eligible applicants include individuals, SMEs, academia and consortia established in EU Member States or eligible Associated Countries, projects must focus on civil applications and sensitive datasets (e.g., CHAIMELEON CT scans) are accessible only via a secure processing environment. Applications are submitted via the F6S platform and the deadline for Challenges 3 and 4 is 8 September 2026, 17:00 CEST.

What is funded

Two generative AI challenges (clinical imaging and automated‑driving scenarios)

A multi-challenge open innovation competition to develop Generative AI prototypes that progress from TRL 2-3 to TRL 4-5. Two featured challenges: 1) Generative AI to enhance clinical imaging datasets (led by EUCAIM) and 2) Generative AI to generate simulation test cases from crash databases and standards (led by Siemens Industry Software with RobustifAI). Support includes technical mentoring, access to secure data/infrastructure (where applicable), and optional HPC access.

Who can apply:Individuals or teams from academia and industry (students 18+, researchers, developers, SMEs, start-ups, spin-offs, NGOs, consortia) established in EU Member States or eligible associated countries; must have technical capacity and AI experience and propose civil applications.

1. 1Two-phase competition: Spark (concept note) and Advance (five-month development)
2. 2Access and evaluation under challenge-specific data and ethical rules (secure processing environments for sensitive data)
3. 3Application via F6S; follow AI-BOOST guidelines and Challenge Owner instructions

Key deadlines:challenges 3 and 4 close 8 September 2026, 17:00 CEST. Apply via the F6S portal and the AI-BOOST competition page AI-BOOST competition and submit via F6S F6S application ¹.

Footnotes

1. 1Primary source: AI-BOOST / EU Funding & Tenders Portal opportunity HORIZON-CL4 and AI-BOOST challenge documents (application and prize structure described on the AI-BOOST website and F6S listing).

Call and Key Dates

Opportunity Title:AI Challenge Competition. Call Title: AI Challenge Competition. Framework / Identifier: AI-BOOST, funded under GA No 101135737, linked to HORIZON-CL4. Deadline for Challenge 3 and 4: 8 September 2026, 17:00 CEST. Spark Phase and Advance Phase structure with final live event in Brussels February 2027.

What this competition is

AI-BOOST runs an open AI Challenge Competition with four targeted Generative AI challenges. The published material describes two specific challenges in detail relevant to this entry: Challenge 3 Generative AI for Enhancement of Clinical Datasets (led by EUCAIM) and Challenge 4 Generative AI for Automatic Test Case Generation from Crash Databases & Standards (led by Siemens Industry Software NV in collaboration with EU RobustifAI). The competition uses a two-phase funnel: a Spark Phase (concept note selection) and an Advance Phase (five-month development, validation, demonstration). Monetary prizes are awarded at both phases and additional special awards are available.

Detailed challenge descriptions (high level)

Challenge 3: Generative AI for Enhancement of Clinical Datasets

Lead organisation:EUropean Federation for CAncer IMages (EUCAIM). Goal: develop generative AI methods to create realistic synthetic cohorts and to augment, complete and harmonise clinical imaging datasets (CT thorax) and associated clinical metadata so as to reduce bias, improve representativeness and support trustworthy medical AI model training and validation. Expected prototype TRL 7 integration into EUCAIM infrastructure for non-exclusive use.

Objectives (Challenge 3):1) Develop a generative AI model capable of producing synthetic cohorts based on the provided data. 2) Identify key demographic, clinical and imaging characteristics as inputs. 3) Use the model to augment datasets and rebalance underrepresented subgroups. 4) Evaluate quality, realism and consistency against original distributions. 5) Demonstrate value for fidelity, bias reduction, completion and harmonisation.

Datasets and data access (Challenge 3):The challenge provides 1088 DICOM Thorax CT imaging studies (baseline timepoint) with an associated JSON file of clinical variables (subject_id; date_baseline_ct; age_at_baseline; gender; tumor_histotype; pd_l1; pd_l1_unknown; local_recurrence_progression; death_related_to_cancer; clinical_staging_date; clinical_stage_group; metastasis_lung; metastasis_pleura; metastasis_lymph_nodes; metastasis_adrenal_gland; metastasis_liver; metastasis_brain; metastasis_bone; metastasis_other; treatment_intent; smoking_status; date_smoking_status; ecog_performance_status; metastasis_clinical_category; progression_recurrence; distant_metastasis_pr; tumor_clinical_category; regional_nodes_clinical_category; death_date). Data originates from CHAIMELEON project (Zenodo record).

Data rights and constraints:datasets are anonymised but sensitive. Access only through a Secure Processing Environment deployed on AI-BOOST resources (no download or personal copies). Applicants must adhere to EUCAIM terms and conditions and dataset-specific terms. Usage requires compliance with the provided EUCAIM usage policy and data transfer agreement instructions. Known limitations include lack of pixel-level annotations and restrictions on data export. Infrastructure support via CINECA secure environment is available.

Evaluation metrics and KPIs (Challenge 3):KPI 1 Synthetic Data Fidelity: average KS distance across continuous variables; average absolute difference in prevalence for categorical variables. Targets: KS ≤ 0.10 for ≥ 80% of variables; prevalence difference ≤ 5% for ≥ 80% of categorical variables. KPI 2 Bias Reduction and Cohort Balancing: reduce imbalance ratio by at least 50% between largest and smallest predefined demographic subgroup. KPI 3 Missing Data Completion and Imaging Harmonization: categorical accuracy ≥ 90%; MAE improvement ≥ 20% vs median imputation baseline; Imaging Completion Rate ≥ 95% for missing CT acquisitions (non-contrast, contrast-enhanced, low-dose); harmonization to reduce kVp intensity distribution differences by ≥ 30% while preserving imaging biomarker stability (median CCC ≥ 0.85 for GLCM Joint Entropy). KPI 4 Robustness: median CCC ≥ 0.85 for GLCM Joint Entropy under Gaussian noise perturbations (σ = 5, 10, 20 HU).

Technical environment and constraints (Challenge 3):Secure Processing Environment: Ubuntu 24.04.4 LTS, Python 3.12.3, PyTorch 2.12.0, TensorFlow / Keras 3.14.1, JupyterLab, dicom2nifti, and standard scientific packages. CINECA can provide secure development environment and optional access to LEONARDO supercomputer. Applicants must process data only inside the secure environment and follow the provided usage guide and EUCAIM terms.

Challenge 4: Generative AI for Automatic Test Case Generation from Crash Databases & Standards

Lead organisation:Siemens Industry Software NV (SISW), collaborating with EU RobustifAI. Goal: convert fragmented European accident and safety evidence into structured, explainable, simulation-ready scenario knowledge to support validation of automated driving systems. The benchmark focuses on extraction of Functional Scenarios, enrichment into Logical Scenarios using multimodal evidence, and matching to reference validation scenario descriptions to assess coverage gaps.

Objectives (Challenge 4):1) Extract recurring Functional Scenarios from heterogeneous accident datasets and cluster them into a common taxonomy. 2) Enrich Functional Scenarios into Logical Scenarios using multimodal evidence (text, images, maps, infrastructure, video). 3) Provide Structured Scenario Representations with attributes, evidence references and confidence metadata. 4) Match participant-generated scenario catalogues to reference validation scenarios (example Euro NCAP subset) to enable validation coverage and gap analysis.

Datasets and resources (Challenge 4):The benchmark may draw on European accident databases, national crash statistics, crash reports, images, videos, maps, infrastructure metadata and open multimodal datasets (representative examples cited include UK Road Safety Open Data, France Accidentologie, KITScenes LongTail, CycleCrash, DeepAccident, CYCLANDS, PREPER, NHTSA CIS, ASAM OpenSCENARIO resources, and reference catalogues like Euro NCAP). Participants may also use other lawful datasets but must comply with access and licensing terms for any non-public sources.

Evaluation metrics and KPIs (Challenge 4):Primary metric dimensions include: quality of functional scenario extraction and taxonomy alignment; completeness and correctness of logical scenario enrichment; robustness of evidence grounding and traceability; accuracy of scenario-to-catalog matching; usefulness for validation coverage assessment. Secondary metrics include explainability, reproducibility, scalability, diversity and representativeness of generated scenarios, and uncertainty representation quality.

Ethical and compliance constraints:participants must consider EU AI Act requirements (transparency, traceability, human oversight, robustness). Explicit disclosure of third-party AI models/APIs/cloud services used is mandatory. It is prohibited to transmit PII or sensitive accident data to external services, to generate physically impossible motions, to hallucinate without uncertainty representation, or to produce inconsistent multi-agent interactions.

Competition structure, eligibility and prizes

Two-phase selection funnel:Spark Phase (concept note), Advance Phase (five-month development and validation). Spark Phase selects five winners per challenge; Advance Phase selects one final winner per challenge. Final live demonstration and pitch event in Brussels (Feb 2027).

1. 1Spark Phase: Submit a Concept Note. Five winners per challenge. Each Spark winner receives €28,500 and advances to Advance Phase.
2. 2Advance Phase: Five-month development and validation programme including a mid-term checkpoint, final algorithm submission, final report, final pitch presentation and live demonstration. One winner per challenge selected for final prize.
3. 3Prizes: Final prize for each challenge winner €100,000. Two Special Awards per challenge: Innovation Excellence Award €12,500 and Responsible AI Award €12,500.

Eligible applicants:Open to individuals and teams from academia and industry. Eligible applicant types explicitly listed: Individuals (students over 18, researchers), SMEs (including startups), universities, research organisations, NGOs, non-profit organisations, and consortia of these. Applicants must be established in EU Member States (including overseas countries and territories) and some Associated Countries (consult the guide for the full list). Projects must focus exclusively on civil applications and applicants must have technical capacity and experience working on AI systems.

Consortium requirement:Participation can be by a single applicant (individual or legal entity) or teams/consortiums. There is no mandatory multi-partner consortium requirement; modular and integrated team proposals are accepted.

Funding type and nature of support:This is a prize-based competition funded under a CSA project. Financial support is delivered as cash prizes (€28,500 for Spark winners, €100,000 final prize per challenge, plus Innovation Excellence and Responsible AI Special Awards of €12,500 each). In addition to monetary prizes, non-monetary support is provided: technical mentorship, access to secure processing environments, optional access to HPC resources (CINECA LEONARDO), domain expertise, integration support, and visibility opportunities.

Budget and duration:The AI-BOOST project budget allocated to the competition is indicated in portal metadata as €535,000. Individual prizes and programme durations: Spark winners enter a five-month Advance Phase development programme. The full Advance Phase is five months; the overall activity from Spark submission to final event spans several months with final live event scheduled February 2027. The project duration referenced in metadata is 8 months for the related CSA activity.

Application process and evaluation

Applications must be submitted via the F6S platform (AI-BOOST challenge pages). Main steps:1) Register on F6S. 2) Complete the application form with all mandatory fields and supporting documents. 3) Submit before the deadline (8 September 2026, 17:00 CEST for Challenges 3 and 4). The portal pages for application are provided by AI-BOOST and F6S; applicants requiring support are instructed to contact support@f6s.com.

Evaluation process and stages:Phase I — SPARK Phase: initial eligibility and concept note evaluation (novelty, alignment with challenge objectives, technical feasibility, innovation potential). Phase II — ADVANCE Phase: five-month development, mid-term checkpoint, final evaluation based on final report, algorithm submission, live demonstration, pitch and audience voting. Evaluation criteria include technical excellence, innovation, scalability, sector relevance, responsible AI principles, sustainability and practical applicability. The evaluation is multi-stage (at least two main stages: Spark and Advance), with intermediate checkpoints.

Application templates and form structure:Applications are submitted through F6S concept note and follow-on application forms. The public materials describe a Concept Note for the Spark Phase and more detailed deliverables for the Advance Phase (final report, algorithm submission, demo, pitch). Typical content expected in the Concept Note includes: problem statement and motivation; proposed approach and novelty; technical approach and tools; datasets and data handling plan (especially for sensitive data); responsible AI considerations; expected outcomes and KPIs alignment; team description, roles and relevant experience; resource needs and computing plan; exploitation and sustainability plan. Applicants must include any declarations required for data usage compliance (e.g., acceptance of EUCAIM data terms if applying to Challenge 3).

Eligibility, geography and compliance

Beneficiary scope (geographic eligibility):Open to applicants established in EU Member States (including overseas countries and territories) and certain Associated Countries. The official guide for applicants lists the full set of eligible Associated Countries; applicants should consult the AI-BOOST guide and the call documentation for the precise list.

Eligible applicant types:Individuals (students, researchers), SMEs (including startups), academia, universities, NGOs, non-profit organisations, research institutes, and consortia composed of these entities. Public sector bodies and large enterprises are not explicitly listed as excluded but the public materials emphasize teams from academia and industry and list SMEs, startups and research organisations as target participants.

Consents and compliance requirements:Applicants must follow data protection regulations (GDPR), the EU AI Act recommendations (transparency, traceability, human oversight, robustness) and any challenge-specific data use terms (e.g., EUCAIM usage policy for Challenge 3). Use of external AI services or LLMs must be disclosed. Participants must avoid sending PII or sensitive data to external services and must document all external dependencies for reproducibility.

Project maturity and targeted TRL

Target project stage:projects are expected to start around TRL 2-3 (research / proof of concept) and progress to TRL 4-5 (validation in relevant environment). Challenge 3 expects a prototype that can reach TRL 7 for integration into EUCAIM infrastructure (the description states TRL 7 as expected outcome for the generative model to be used to improve cohorts). Challenge 4 aims to move solutions from TRL 3 to TRL 5 (validation in relevant environments).

Funding details and co-funding

Funding mechanism:prize awards financed by the AI-BOOST project (a CSA). Winners receive cash prizes; non-winners do not receive project grants. There is no mandatory co-funding requirement described for applicants to receive a prize. The competition provides in-kind support (technical guidance, infrastructure access) but does not require applicants to secure matching funds. Applicants are responsible for their own project costs unless explicitly covered by prize money or support agreements.

Success rates and expected selectivity:Spark Phase: five winners selected per challenge from the pool of eligible submissions. Advance Phase: one final winner per challenge from the five Spark winners. Because exact numbers of applicants are not published, a numeric success rate cannot be precisely stated; the selection funnel implies a selection ratio of 5 winners at Spark per challenge and 1 final winner from those five (i.e., 20% of Spark entrants progress to final winner for each challenge). Overall success rate from open applicants to Spark winner is unknown and depends on the number of applications received.

How to apply and where

Apply via the F6S platform pages for the AI Challenge Competition (links provided on AI-BOOST website). Applicants must register on F6S, complete the online Concept Note/application form and upload required supporting documents before the deadline. For Challenges 3 and 4 use the F6S page at f6s.com. For support contact support@f6s.com. Further public documentation and challenge annexes are available on AI-BOOST website and linked PDFs for Challenge 3 and Challenge 4.

Mentioned countries and geographic references

Explicit geographic references in the opportunity text:European Union (EU) Member States and some Associated Countries. The Challenge 4 description references European accident databases and national datasets (examples provided: UK, France, Belgium, Spain (Barcelona) and pan-European resources). Use of European datasets and emphasis on European validation contexts is central.

Technical and scientific requirements

Applicants should demonstrate technical capacity and experience working on AI systems and generative AI, with domain-specific expertise depending on the challenge (medical imaging, synthetic data generation, multimodal foundation models, retrieval-augmented generation, knowledge graphs, agentic AI, scenario representation and matching, simulation pipelines). Applicants must be able to operate within secure processing environments and follow data governance and privacy rules. For Challenge 3 strong experience with medical imaging, CT data, DICOM handling, radiomics, statistical fidelity metrics and harmonisation methods is required. For Challenge 4, experience with accident databases, multimodal evidence extraction, scenario taxonomies, scenario formalisms (OpenSCENARIO), explainability and traceability is required.

Application stages and deliverables

1. 1Stage 1: Spark Phase — Submit Concept Note. Deliverables: concept note and supporting documents. Evaluation: eligibility and concept scoring. Outcome: selection of 5 Spark winners per challenge.
2. 2Stage 2: Advance Phase — Five-month development and validation. Deliverables: periodic progress updates, mid-term checkpoint deliverable, final algorithm submission, final report, final pitch presentation and live demonstration. Evaluation: technical performance against KPIs, final demonstration and audience/evaluator scoring. Outcome: one final winner per challenge and special awards.

Summary and guidance for applicants

AI-BOOST AI Challenge Competition invites individuals, startups, SMEs, academic teams and research organisations established in eligible EU and Associated Countries to compete in focused generative AI challenges. Challenge 3 targets trustworthy synthetic data generation and dataset augmentation for medical imaging cohorts with strict data governance inside a secure processing environment. Challenge 4 targets multimodal generative AI to convert fragmented accident and safety evidence into structured, traceable scenario knowledge for automated driving validation. The funding is prize-based with clear KPIs and evaluation protocols. Applicants should prepare a strong concept note that addresses the challenge objectives, demonstrates technical feasibility and responsible AI practices, and documents data handling and reproducibility plans. Selected Spark winners receive cash to support development and access to mentorship and infrastructure; Advance Phase winners receive a larger deployment-oriented prize and visibility at a final live event.

Quick action checklist for applicants:1) Confirm eligibility (establishment in EU or eligible Associated Country; team composition acceptable). 2) Read challenge annexes and data usage policy (EUCAIM terms for Challenge 3). 3) Register on F6S and prepare Concept Note addressing objectives, KPIs, datasets and technical approach. 4) Prepare declarations on data handling, AI dependencies and compliance with EU AI Act considerations. 5) Submit via F6S before 8 September 2026 17:00 CEST for Challenges 3 and 4. 6) If selected, be prepared for a five-month development programme including mid-term checkpoint and final live demo.

For full challenge annexes and data resources consult the AI-BOOST challenge pages and the published PDFs for Challenge 3 and Challenge 4. The public competition pages and application platform (F6S) host detailed guidance and application forms AI-BOOST Challenge Page F6S apply (Challenges 3 & 4). ¹

Footnotes

1. 1Primary source material: AI-BOOST project pages and challenge annex PDFs linked from AI-BOOST website and the EU Funding & Tenders Portal for the AI-BOOST call (Open innovation: Addressing Grand challenges in AI (AI Data and Robotics Partnership) (CSA)).

Competition Overview and Phases

The AI Challenge Competition is a two-phase open innovation event funded by the European Union under GA No 101135737, targeting teams from academia and industry to develop breakthrough Generative AI solutions ¹. The competition consists of a Spark Phase for concept note submission, awarding five winners per challenge with €28,500 each, followed by an Advance Phase involving a five-month development programme culminating in a live demonstration in Brussels in February 2027 ². One final winner per challenge will be selected to receive a €100,000 prize, with two special awards (Innovation Excellence and Responsible AI) of €12,500 each also available per challenge ³.

Challenge 3: Generative AI for Enhancement of Clinical Datasets

Challenge Owner:Led by EUropean Federation for CAncer IMages (EUCAIM), part of the European Cancer Imaging Initiative ⁴.

This challenge focuses on using Generative AI to create realistic synthetic patient cohorts to enhance the quality and representativeness of incomplete or imbalanced clinical imaging datasets, specifically Thorax CT scans from 1,088 subjects ⁵. The solution must identify key demographic and clinical characteristics to ensure synthetic data remains realistic, aiming to reduce bias and improve fairness in medical research and clinical decision-making ⁶.

1. 1Develop a generative AI model to generate synthetic cohorts based on provided data
2. 2Identify key demographic, clinical, and imaging characteristics for model input
3. 3Augment existing datasets to improve balance across population subgroups
4. 4Evaluate quality, realism, and consistency of generated data against original distributions
5. 5Demonstrate added value in fidelity, bias reduction, and harmonization metrics

Challenge 4: Generative AI for Automatic Test Case Generation

Challenge Owner:Led by Siemens Industry Software NV with collaboration from EU RobustifAI ⁷.

This challenge aims to enhance the safety and validation of autonomous driving systems by automating the creation of simulation scenarios using Generative AI ⁸. Participants must transform fragmented accident reports, visual data, and international safety standards into structured, simulation-ready formats to replace slow manual processes and identify critical safety gaps ⁹.

1. 1Extract Functional Scenarios from heterogeneous accident datasets and harmonize them
2. 2Enrich Functional Scenarios into Logical Scenarios using multimodal evidence (text, images, maps)
3. 3Package outputs into a common Structured Scenario Representation with traceability
4. 4Match accident-derived scenarios to reference validation catalogs (e.g., Euro NCAP) to assess coverage
5. 5Identify meaningful coverage gaps and prioritize insights for simulation-based validation

Eligibility and Application Details

Eligible applicants include individuals (students, researchers), SMEs (including startups), academia, universities, NGOs, and consortiums established in EU Member States or Associated Countries ¹⁰. Participants must have technical capacity and experience working on AI systems, with projects focusing exclusively on civil applications ¹¹.

1. 1Registration via the F6S platform (f6s.com)
2. 2Dully complete the application form (all mandatory fields must be completed)
3. 3Submit the application form before the deadline: 8 September 2026, 17:00 CEST
4. 4Prepare required supporting documents and submit via F6S

The competition offers technical guidance, access to challenge-specific resources, optional access to CINECA’s LEONARDO supercomputer, and opportunities to establish relationships with Challenge Owners ¹². An info webinar is scheduled for 22 July 2026 at 11:00 CEST to explain the funding structure and evaluation criteria ¹³.

Footnotes

1. 1Official competition details: AI Challenge Competition
2. 2Phase structure and prizes: AI-BOOST Competition Phases
3. 3Special awards details: AI-BOOST Awards
4. 4Challenge Owner EUCAIM: EUCAIM Website
5. 5Challenge 3 Description: Generative AI for Clinical Datasets PDF
6. 6Dataset details: CHAIMELEON Dataset
7. 7Challenge Owner Siemens: Siemens Industry Software
8. 8Challenge 4 Description: Generative AI for Test Cases PDF
9. 9Data sources for Challenge 4: European Accident Databases
10. 10Eligibility criteria: AI-BOOST Eligibility
11. 11Civil application focus: AI-BOOST Guidelines
12. 12Program benefits: AI-BOOST Benefits
13. 13Info Webinar: Webinar Registration