NASK Researchers Win the Warsaw Edition of the CISPA European AI & Cybersecurity Hackathon Championship

The winning team, Syntax Terror, consisted of Mateusz Kowalczyk and Aleksander Szymczyk from the Artificial Intelligence Security and Transparency Department at NASK, along with two individuals from the National Centre for Nuclear Research (NCBJ); thanks to this, they secured a place in the final, which will take place in Germany

The hackathon focused on practical challenges related to the security and trustworthiness of artificial intelligence systems. As part of the competition, participants tackled tasks such as:

• detecting whether a text was generated by a language model using so-called digital watermarks (LLM Watermark Detection),
• analyzing how multimodal models may “memorize” sensitive data used during training and whether such data can be partially reconstructed despite prior masking (e.g., phone numbers, email addresses, or payment card details) (Multi-modal Memorization in LMMs),
• estimating the extent to which specific datasets were used during the training of AI models (Dataset Usage Cardinality Inference).

The tasks were highly demanding and practical in nature, which further highlights the significance of this achievement. The team’s success also demonstrates effective collaboration between NASK and NCBJ in the field of AI security research.

More information is available in the LinkedIn post published by CISPA.

Congratulations to the winners, and we wish them the best of luck in the finals!