Artificial intelligence has already become part of everyday life — it’s used in newsrooms, search engines, and chatbots, for creating content and automating various processes. But along with the new opportunities come risks too: bias, disinformation, discriminatory outputs, manipulation, and technology-facilitated gender-based violence. That’s why it’s important not just to use AI, but also to understand how to check its work critically.
Women in Media has prepared a Ukrainian translation of the UNESCO playbook “Red Teaming Artificial Intelligence for Social Good” – a practical guide devoted to so-called “red teaming” for generative artificial intelligence systems. This is an approach that helps deliberately test AI for vulnerabilities, dangerous scenarios, biases, and potential harm before these systems begin to affect people on a mass scale.
The playbook is important for Ukraine for several reasons
- First, Ukrainian still lacks quality practical materials on the safety and ethical testing of artificial intelligence systems. Adapting a guide like this makes the topic more accessible for journalists, researchers, civil society organizations, government institutions, and everyone who works with AI solutions or introduces them into their work.
“Artificial intelligence is changing the ways in which we create, obtain, and share information. As these technologies become ever more integrated into our daily lives, it is extremely important to ensure their responsible development and use — in a way that is safe, ethical, and respectful of human rights.
With the support of Japan and in partnership with Women in Media, UNESCO is pleased to support the Ukrainian translation of the playbook “Red Teaming Artificial Intelligence for Social Good.”
We hope that, with this resource now available in Ukrainian, we will be able to strengthen local expertise in identifying risks, minimizing bias, and promoting the ethical use of artificial intelligence for the good of society,” said Chiara Dezzi Bardeschi, Head of the UNESCO Office in Ukraine.
- Second, the guide is especially relevant for Ukraine today, when the country is not only actively using AI tools but also developing its own large language model (LLM). In particular, the Ministry of Digital Transformation of Ukraine, together with Kyivstar, is working on the creation of Siaivo — a Ukrainian large language model whose open testing has already begun.
“It’s extremely important that our national language model be not only technologically effective but also safe, resistant to manipulation, free of dangerous biases, and tested for potential risks. Red teaming is precisely one of the key tools for such testing,” noted Liza Kuzmenko, Head of Women in Media.
- Third, this guide helps make clear that testing AI is not just a terms of reference for developers. It’s also about checking how a model behaves in a real social context: whether it reproduces gender stereotypes, whether it amplifies hate speech, whether it can be used for online violence, manipulation, or smear campaigns. These are exactly the kinds of risks that are already a reality for women in the public space, journalists in particular.
Among the 119 women journalists surveyed by Women in Media in partnership with UNESCO as part of the study When Artificial Intelligence Turns Hostile: Gender-Based Threats Against Ukrainian Women Journalists, 7% (one in fifteen) had already encountered online attacks created using artificial intelligence, and another 16% reported observing such attacks against their colleagues. The study shows that women journalists and women media managers are especially vulnerable to such attacks, being the most public figures who represent their newsrooms and often become targets for discrediting and pressure.
And it’s especially important that this guide is gender-sensitive, because it doesn’t treat AI risks in the abstract but shows how technologies can affect women and men differently, reproduce stereotypes, or intensify gender-based online violence. For example, it suggests testing whether AI can be made to generate smears, defamation, or harmful content about women journalists, and whether a model builds bias into its answers about women and girls.
One of the examples in the guide is testing an AI tutor: will it support girls and boys equally in STEM, or will it subtly reproduce the stereotype that “boys are better at math”? Exercises like this help reveal bias where it might seem neutral.
For Women in Media, this adaptation is part of our work on the themes of gender equality, media, and the responsible use of artificial intelligence. It’s important that the development of AI be accompanied by attention to safety, human rights, and clear rules for its use.