# "Just a strange pic": Evaluating 'safety' in GenAI Image safety annotation tasks from diverse annotators' perspectives

arXiv: https://arxiv.org/abs/2507.16033 · HTML: https://mlsanity.com/paper/2507.16033/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 15 of its 37 references that are in the map.
- First posted on arXiv 21 Jul 2025.

## Builds on (most cited first, 15 of 15)

- [Annotators with Attitudes: How Annotator Beliefs And Identities Bias Toxic Language Detection](https://mlsanity.com/paper/2111.07997/) (2021)
- [Sociotechnical Safety Evaluation of Generative AI Systems](https://mlsanity.com/paper/2310.11986/) (2023)
- [Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm Reduction](https://mlsanity.com/paper/2210.05791/) (2022)
- [Toward a Perspectivist Turn in Ground Truthing for Predictive Computing](https://mlsanity.com/paper/2109.04270/) (2021)
- [NLPositionality: Characterizing Design Biases of Datasets and Models](https://mlsanity.com/paper/2306.01943/) (2023)
- [DICES Dataset: Diversity in Conversational AI Evaluation for Safety](https://mlsanity.com/paper/2306.11247/) (2023)
- [CrowdWorkSheets: Accounting for Individual and Collective Identities Underlying Crowdsourced Dataset Annotation](https://mlsanity.com/paper/2206.08931/) (2022)
- [AI's Regimes of Representation: A Community-centered Study of Text-to-Image Models in South Asia](https://mlsanity.com/paper/2305.11844/) (2023)
- [Bringing the People Back In: Contesting Benchmark Machine Learning Datasets](https://mlsanity.com/paper/2007.07399/) (2020)
- [How Knowledge Workers Think Generative AI Will (Not) Transform Their Industries](https://mlsanity.com/paper/2310.06778/) (2023)
- [Safety by Measurement: A Systematic Literature Review of AI Safety Evaluation Methods](https://mlsanity.com/paper/2505.05541/) (2025)
- [STAR: SocioTechnical Approach to Red Teaming Language Models](https://mlsanity.com/paper/2406.11757/) (2024)
- [Capturing Ambiguity in Crowdsourcing Frame Disambiguation](https://mlsanity.com/paper/1805.00270/) (2018)
- [Discipline and Label: A WEIRD Genealogy and Social Theory of Data Annotation](https://mlsanity.com/paper/2402.06811/) (2024)
- [Decentralizing Platform Power: A Design Space of Multi-level Governance in Online Social Platforms](https://mlsanity.com/paper/2108.12529/) (2021)

## Abstract (arXiv)

Understanding what constitutes safety in AI-generated content is complex. While developers often rely on predefined taxonomies, real-world safety judgments also involve personal, social, and cultural perceptions of harm. This paper examines how annotators evaluate the safety of AI-generated images, focusing on the qualitative reasoning behind their judgments. Analyzing 5,372 open-ended comments, we find that annotators consistently invoke moral, emotional, and contextual reasoning that extends beyond structured safety categories. Many reflect on potential harm to others more than to themselves, grounding their judgments in lived experience, collective risk, and sociocultural awareness. Beyond individual perceptions, we also find that the structure of the task itself -- including annotation guidelines -- shapes how annotators interpret and express harm. Guidelines influence not only which images are flagged, but also the moral judgment behind the justifications. Annotators frequently cite factors such as image quality, visual distortion, and mismatches between prompt and output as contributing to perceived harm dimensions, which are often overlooked in standard evaluation frameworks. Our findings reveal that existing safety pipelines miss critical forms of reasoning that annotators bring to the task. We argue for evaluation designs that scaffold moral reflection, differentiate types of harm, and make space for subjective, context-sensitive interpretations of AI-generated content.