Generative AI has created unprecedented opportunities for interacting with ancient sources (Assael et al. 2025; Sommerschield et al. 2023). The ability to query, summarise, or translate specialised material using pre-trained or general-purpose models has rapidly expanded engagement: students, researchers, and non-specialists can now access materials that once required years of linguistic training (Assael et al. 2025; Peddar 2025; Sommerschield et al. 2023). Yet the widespread availability of AI tools contrasts sharply with the limited number of specialists who understand how these systems are constructed, how their training data shape their behaviours, and where their fundamental limitations lie (Mak / Luo 2025). This gap creates conditions in which errors or distortions readily propagate, often unnoticed.

For ancient low-resourced languages, these risks are magnified. The corpora on which these languages survive are inherently “small data”: broken tablets, damaged inscriptions, and disparate manuscripts. Their scripts (cuneiform, hieroglyphic, hieratic, demotic, sinographic, kana/hanja systems) are structurally complex, polyvalent, and highly context-dependent. Their semantic fields are embedded in cultural, religious, and historical frameworks that models cannot infer from predominantly modern Western training data. As a result, AI outputs frequently flatten nuance, erase ambiguity, and impose linguistic or cultural norms unsuited to the material (Blasi et al. 2022).

This mini-conference directly addresses DH2026’s theme of Engagement by scrutinising the tension between expanding accessibility and maintaining interpretive rigor. Engagement-driven uses of AI, such as translation for public audiences, summarisation for classroom use, or automated paraphrasing for introductory materials, may unintentionally reinforce oversimplification or revive outdated scholarly assumptions. When LLMs impose modern Mandarin grammar onto Old Chinese or collapse the polysemy of Ancient Egyptian and Akkadian lexemes, they do more than make mistakes: they generate interpretive narratives that appear coherent, authoritative, and accessible, while masking the underlying instability of the evidence. Such distortions directly affect pedagogy and research, shaping how ancient cultures are understood, represented and ultimately taught and learnt.

The conference considers these risks not as arguments against AI, but as invitations to think critically about engaged scholarship. It asks:

• How can scholars promote accessibility without compromising methodological principles?
• How can emerging technologies support, rather than erode, the interpretive traditions of philology, epigraphy, and historical linguistics?
• Can multilingual scholarly communities collaborate to set discipline-specific boundaries and best practices?
• What forms of digital literacy are necessary for researchers, educators, and students to engage responsibly with AI?

This event gathers experts across diverse ancient-language traditions to highlight structural vulnerabilities shared across fields, while remaining attentive to cultural and linguistic specificity, and to build a shared understanding of responsible AI adoption within Ancient World Studies. Through a series of presentations, collaborative discussions, and brainstorming sessions, this mini-conference aims to:

• Develop a clearer understanding of the methodological vulnerabilities of low-resourced ancient-language corpora in AI contexts;
• Foster field-specific awareness of how LLMs may distort primary and secondary sources;
• Identify strategies for balancing accessibility with rigor in teaching, research dissemination, and public outreach;
• Produce a collaboratively authored set of responsible use guidelines to support transparent, critical, and disciplined engagement with AI in ancient languages.

The six-hour-long event includes an opening session, a keynote presentation, and three main presentation sessions, each followed by a discussion, around the following themes:

Session 1:

AI and Translations: Interpreting Uncertain and Ambiguous Data (On interpretative behaviour of AI toward culturally specific, fragmentary, and ambiguous texts whose translations allow for multiple possible rendering);

Session 2:

AI and Graphic Elements: Innovation or Illusion in Low-Resource Contexts? (On advantages and risks of using AI to recognize, interpret and reconstruct graphic elements and palaeography);

Session 3:

AI and Philological and Educational Practices: Interpretative Assistant or Source of Dependence? (On positive and negative impacts of AI for textual analysis, teaching and learning).

The final part of the event includes two collaborative sessions:

Session 4:

Moderated discussion of the main themes and ideas presented throughout the event to identify areas of risk and best practices of AI use in teaching and research.

Session 5:

Collaborative drafting of a guidelines document to support responsible use of AI in teaching and research. To ensure that this objective is achieved within the session, provisional guidelines are drafted by the conference organizers and circulated in advance among speakers and other specialists for feedback and revision, so that a first draft will be available on the day of the event.

The event includes the following contributions:

Yael Assouline (Ariel University, Israel): "Scaling the Detection of Historiolae in the Papyri Graecae Magicae: A Corpus Annotation Project Using LLMs and Model Context Protocols";

Greg Baker, Shirley Chan, Vanessa Enriquez Raido, Greta Hawes (Macquarie University): "Into the Parallage: harnessing abundance, plurality and divergence in AI";

Doaa Elalfy (Hebrew University of Jerusalem): "Imposing Coherence: AI and the Construction of Meaning in Greek–Arabic Texts";

Lee Gunhyuk (Central European University, Vienna): "Can AI Read Coptic Ostraca?: Formula, Fragment, and the Reconstruction of Meaning in the O.Frange 1–27";

Eva Maria Hemauer (HU Berlin/LMU Munich): "Can AI Read Your Hieroglyphs? Evaluating LLM Performance across Ancient Egyptian Script Traditions, Transliteration Practices, and Low-Resource Ancient Languages";

Alexandre A Loktionov (HSE University, Moscow/University of Cambridge), Innokentiy Humonen (AXXX, Moscow), Maksim Golyadkin (HSE University, Moscow), Valeria Rubanova (ITMO University, St. Petersburg), Ekaterina Alexandrova (HSE University, Moscow), Ilya Makarov (AXXX, Moscow/Trusted AI Center, Russian Academy of Sciences, Moscow): "AI in the Hieroglyphic Classroom? A Pilot Study of Human-in-the-Loop Assistance, Cognitive Load, and Pedagogical Integration";

Celia López Castillo (University of Sevilla), David Galindo Diez (University of Zaragoza): "Generative AI for the Interpretation of Non-Alphabetic Visual Systems: A Cross-Cultural Study of East Asian Scripts and Iconography";

Isabelle Marthot-Santaniello (University of Basel): "AI for the paleography of Ancient Greek Papyri: some reflections from eight years of collaboration between Computer Vision and Papyrology";

Pavla Rosenstein (Yale University): “Learn Hammurabi: Utilizing ‘Vibe Coding’ for Pedagogy and Public Outreach“;

Shuning Zhang (University of Tsukuba): "The Role of AI in Philological Research: A Case Study of a Bilingual Manchu–Classical Chinese Database".