desegma

Desegma-IT - Detection and Segmentation of Machine Generated Text in Italian - is a Shared Task of the 9th evaluation campaign organized by the EVALITA initiative.

Recent advancements in Generative AI and Large Language Models (LLMs) have led to the development of systems, such as GPT-series, Claude, Llama and DeepSeek among others, that can generate text that is often indistinguishable from human-written content. This capability, along with the many beneficial applications of LLMs, also enables malicious actors to generate synthetic content for deceptive purposes. For example, it can be used to manipulate online traffic and spread misinformation through content farms or to influence human revisions of sensitive documents in critical domains, such as scientific peer review.

Beyond public opinion concerns, the European Commission will enforce compliance with the AI Act. In particular, article 50 points 2 and 4 require that text generated by General Purpose AI (GPAI) have to be marked and automatically recognizable, making the need for machine generated text detectors, not only desirable but required by law. Therefore, to comply with the AI Act and mitigate concerns about the undisclosed use of machine-generated text (MGT), the development of reliable detection methods is essential for a responsible deployment of generative AI.

Desegma-IT’s shared tasks of Detection and Segmentation aim to stress-test the robustness of machine-generated text detectors by evaluating their performance under settings where the I.I.D. (independent and identically distributed) assumption does not hold. While state-of-the-art MGT detectors have reported high accuracy, such results often stem from unrealistic experimental settings: for example, relying on prior knowledge of the text generator, or failing to consider domain shifts and efficient fine-tuning - or post-tuning - strategies.

Important Dates

• Data Release: September 22nd, 2025
• Evaluation Window: dates will be announced by October 13th, 2025
• Camera Ready Deadline: February 2nd, 2026
• Workshop Date: February 26th-27th, 2026, Bari (Italy)