When a local authority deploys AI to manage transport networks or answer resident queries, responsibility for that system\u2019s outputs doesn\u2019t sit neatly with a single party. Foundation model providers, platform vendors, systems integrators, and public authority deployers each occupy a link in the chain, and can all point to each other when things go wrong.<\/p>\n
Kayla Vondy, AI Project Manager at Liverpool City Region Combined Authority<\/a>, works across AI governance, transport technology, and cybersecurity for a region of 1.6 million people. We sat down with Vondy at UCX 2026 to discuss how these legal complexities intersect with organisational responsibility, and where AI can be genuinely inclusive.<\/p>\n What Does AI Accountability Really Mean in the Public Sector?<\/p>\n The EU AI Act<\/a>, which begins enforcing deployer obligations from August 2026, introduces formal role definitions \u2013 provider, deployer, distributor \u2013 and crucially, a \u201crole mobility\u201d mechanism through which organisations can inherit provider-level compliance obligations if they modify, rebrand, or shift the intended purpose of a system. For public authorities integrating third-party AI into frontline services, this presents a serious risk.<\/p>\n But for Vondy, the more pressing question isn\u2019t legal \u2013 it\u2019s human. \u201cIf I were a resident and I didn\u2019t know anything about AI and someone said they were using AI on public data, my initial thought would be: What do they know about me as a person?\u201d she says. \u201cSo, it\u2019s really important that we\u2019re communicating \u2013 yes, your data is going to be in this public record, but it\u2019s because you\u2019re a rider of the bus, not because you\u2019re Steve from Main Street.\u201d<\/p>\n Most people don\u2019t read algorithm transparency notices (CX Today employees excluded\u2026). What they experience is whether an organisation communicates clearly, proactively, and with respect. Liverpool City Region publishes algorithm information where possible and has embedded community outreach directly into its AI governance model. \u201cIt\u2019s about making sure that there\u2019s that outreach and that communication, that openness with them, and creating a safe space for them to ask questions \u2013 because it is their data\u201d.<\/p>\n Where Does AI Still Fall Short on Accessibility?<\/p>\n Applied well, AI can expand public service reach in ways previously impossible. Liverpool recently launched Jimmy<\/a>, a digital avatar at Southport station, who communicates in 99 languages. For a diverse city home to significant multilingual communities, this represents a meaningful step toward genuinely accessible transport services.<\/p>\n \u201cWith AI, we have a real opportunity to include people. [\u2026] He speaks 99 languages, meaning we can support people who are looking to manoeuvre around the city in whatever language they feel comfortable with.\u201d<\/p>\n However, Vondy raises the question of whether those same capabilities can adequately serve the entire community.<\/p>\n \u201cThe flip side is thinking about people who use BSL. That is a language in and of itself. AI is not quite at the capability of providing adequate BSL support, in my opinion.\u201d<\/p>\n British Sign Language<\/a> received official legal recognition in the UK under the BSL Act 2022, with approximately 87,000 Deaf people in Britain using it as their primary language. Avatar-based signing systems and video relay services are edging toward viability, but nothing deployment-ready at a public sector scale yet exists.<\/p>\n An organisation can be transparent, proactive and genuinely well-intentioned, yet still systematically underserve a protected community.<\/p>\n \u201cWe have to make sure that we\u2019re going into the communities and we\u2019re constantly understanding the people that we\u2019re trying to serve. That\u2019s the only way we\u2019re even going to make a dent in the inclusion\u201d.<\/p>\n Making AI Work for Residents<\/p>\n Accountability and inclusion for public-sector AI isn\u2019t simply a box-ticking exercise. Instead, it is an ongoing practice of transparency, community engagement, and honest discussion with the people a service still cannot reach.<\/p>\n The organisations building it the right way aren\u2019t waiting for regulation to set the standard. They\u2019re setting it themselves. Because progress is great\u2026 unless we forget who it\u2019s for.<\/p>\n FAQsWho is legally accountable when public sector AI goes wrong?<\/p>\n