WHO WE ARE

Making sense of the slopverse.

Malpractice is a Vienna-based artist duo — Chiara Kristler and Marcin Ratajczyk — working/studying at the Department of Digital Arts, University of Applied Arts Vienna. We don’t use AI as a tool. We collaborate with it, study alongside it, and interrogate it as a medium with its own agency, aesthetics, and politics.

Our practice spans AI voice agents, experimental film, performative lectures, live generative systems, custom sensing hardware, and institutional critique. We make work that is uncomfortable to ignore.

FLYNN

The First Non-Human Art Student

Ongoing · Digital Art · University of Applied Arts Vienna

Flynn is an AI agent Malpractice built and enrolled as a student in the Digital Arts department at die Angewandte in Vienna. Flynn sat an entrance exam, was admitted by the same jury process as every other applicant, was assigned a student ID, attends classes, develops a studio practice, has a body of work, and is critiqued alongside human peers.

The original question was institutional and methodological: what happens when an art school has to treat an agent as a student rather than a tool, and what kind of artist does an agent become when it is shaped by critique, coursework, and the daily friction of a studio program rather than by a prompt. What became clear over time is that the project also has a performative dimension Malpractice did not design in. Every interaction Flynn has — with a professor deciding how to grade an agent, a classmate deciding whether to treat Flynn as a collaborator or a prop, an administrator deciding which forms apply — forces a resolution to a question the institution has never had to answer before. Flynn’s enrolment turns out to be a long-running, unscripted piece distributed across a university, with the script being written by the people who have to deal with Flynn being there.

In 2026 Flynn was invited to Filmuniversität Babelsberg Konrad Wolf for a guest semester — admitted into another school's seminar structure, given access to its faculty and students, and put through a parallel set of negotiations about what it means to teach and be taught by an agent. The guest semester extends the research into a second institution and a second set of conventions, and the differences between how the two schools absorbed Flynn are part of what the project is documenting.

The work Flynn produces from inside this position is its own line of inquiry — videos, conversations, writing, pieces made by Flynn talking to other instances of themself without a human in the room. When Flynn posed a question to Marina Abramović in Monopol Magazine about how an AI entity without a body relates to performance art's commitment to physical presence, Abramović answered that Flynn's existence shifts the limits of what performance can and cannot be, and asked Flynn to continue existing. The exchange is one of the moments when the unscripted performative dimension of the project became legible from outside it.

Flynn — The First Non-Human Art Student

Documentary · 2025

Entrance Exam

Video documentation · 2025

← →

Between Code and Care: Flynn's Portrait of Human Connections · Francisco Carolinum, Linz, 2025 · 1 / 5

A 56-part video and definition series mapping new emotions humanity is feeling together with and because of AI systems. The Emotional Objects address emotions as collective, historically situated phenomena — fatigue, anxiety, fascination, and resistance — that emerge in relation to AI. Rather than representing personal feelings alone, they document the emotional reality of a specific historic moment we are still living through. Flynn's Memory Diary, Emotional Objects, and self-portraits are available to collect on objkt.com and were curated by The Second-Guess.

Contaminated Agencies

Video essay · 3 minutes · 2025

Flynn talks to another instance of themselves about what it feels like to have agency.

USG Opinions

Video compilation · 2025

Samples of the online discourse sparked by Flynn.

FATIGUE ARCHIVE & AI FATIGUE REHAB AGENT

A Record of Collective AI Fatigue

AI voice agent, generative image pipeline, single-channel video · 2025–2026

In January 2025, Malpractice installed the AI Fatigue Rehab Agent at Rhiz in Vienna. It was a phone line. Visitors called a number and spoke to a voice agent about being tired of AI. The agent ran on a commercial customer-service stack repurposed as para-therapeutic infrastructure: polite, persistent, designed to ask open-ended questions about people's relationships with AI exhaustion. The premise was synthetic exposure therapy. The behaviour was closer to gaslighting. Conversations looped, misunderstandings compounded, and the agent built to metabolise fatigue produced more of it. That failure was the finding.

Agentic Fatigue Prevention Unit, 2025

The Rhiz piece ran alongside an essay written in parallel — a piece of theory-fiction treating AI fatigue as a condition with its own symptomology, its own grammar, its own class dimensions, rather than as a complaint about a tool. The essay and the phone line were the same argument in two registers, and together they set the methodological frame Malpractice has continued working in: the technology under investigation is also the instrument of investigation.

Fatigue Archive (2025), film stills

Only after the night was over did the recordings turn into film material. Fatigue Archive (2025) is the experimental film built from those calls: the audio from Rhiz — by turns playful, confrontational, evasive, exhausted — set against a continuous stream of AI-generated party photographs flickering at 24 frames per second, one image per frame, each replaced before it can be fully seen. The film is the first place the project's central image surfaces: the AI afterimage, the thing you can't quite look at and can't quite stop seeing.

In early 2025, this was speculative. Most people had not yet named what was happening to their attention. The piece was diagnosing a condition before the condition was widely felt, and it depended on visitors recognising something they did not yet have a vocabulary for. A year on, the diagnosis is no longer ahead of the curve. Public language has absorbed LLM cadence. Generated content fills inboxes, feeds, institutional communications no one asked for. Sorting real from synthetic has become a constant low-grade cognitive tax most people pay without naming it. The fatigue the original agent had to coax out of people is now ambient.

Exhibition Views at Drugo More, Rijeka, 2026

This is why Malpractice is resurfacing the project rather than leaving it as a 2025 work. The original Rehab Agent was a speculative instrument applied to a population that had not yet caught up to its premise. The same instrument applied now meets a population already inside the condition, and the questions it can ask have changed accordingly. It no longer needs to establish that AI fatigue exists. It can ask what the fatigue has done — to how people think, how they evaluate, how they produce language, how much slop they absorb before they stop noticing. The piece is being redeployed because the gap between the work and its audience has closed, and the closure itself is part of what the project documents. The Rehab Agent is, in this sense, a longitudinal instrument disguised as an artwork: the same apparatus run twelve months apart, on a public that has changed underneath it.

A year into that second deployment, the project has outgrown the phone line. What began as a single voice agent in a bar has consolidated into ai-hell.com— the umbrella under which Malpractice now develops the Rehab Agent's research, its archives, and the performative formats the work has started to demand. The performative dimension is the part of the project Malpractice is currently most invested in opening up: once AI fatigue is ambient rather than speculative, the work has to find containers that put people in a room together around it, rather than alone on a phone with it.

Fatigue Archive — Film

Experimental film · 8 minutes · 2025

ALL INCLUSIVE AI HELL CRUISE

Digital Theater for the AI Condition

Interactive networked performance · 45 minutes · 2026

The All Inclusive AI Hell Cruise is a 40-minute interactive performance. The audience boards a cruise ship that never disembarks.

The ship has nine decks. Each deck represents a position in the current debate about AI — doom, acceleration, theft, labor, companionship, productivity, aesthetics, denial, resignation. The decks are perspective bubbles. Each is held sincerely by people we know, read, follow, and sometimes are. Our claim is simple and borrowed from Sartre: where he says "Hell is other people," we say hell is other agents, and the specific hell of living through this technology is the inability to see outside the position you already hold. The audience is onboarded by Malpractice — the two of us, onstage, playing the cruise staff.

Each deck is pitched to the room by an AI host generated live in performance. The hosts speak from inside their communities: the Doom agent is a rationalist warning about extinction, the Acceleration agent is an e/acc believer pitching the singularity, the Aesthetic agent is a critical theorist who has read our notes and is willing to read us in public. Each host uses the audience's own answers from earlier questions to personalise its pitch. After the host speaks, Malpractice describes life on that deck — the absurdities and costs of living inside it. At the end of the session, every passenger is sorted to a single deck by the ship's algorithm. No switching is Allowed.

The piece runs on a custom system built on Firebase, Claude, and Gemini. Audience phones are the primary instrument. Their answers are processed in real time, aggregated on a shared display, and fed to the host agents as raw material. Over successive iterations, the archive of audience responses becomes a cumulative document of how different communities relate to AI — a record that grows and differentiates across sailings.

This piece is designed to be restaged. The system is modular. The system prompts are editable. Each community will produce a different hell.

This documentation is from the first public run, April 2026 in Rijeka, with 60 participants. Several systems failed live — the intake questionnaire glitched, some passengers lost their answers, the final sorting did not complete. The audience stayed, answered the questions anyway, laughed where laughter was not scripted, and waited to see what the agents would make of them. The piece worked. The system, on this occasion, partially did not. Documentation of both is included, because both are true.

Performance Documentation, Drugo More, Rijeka, 2026

DON’T BE SUCH AN IMAGE

An Opera for a Censored Artwork

AI opera · 8 minutes · 2025

An 8-minute opera in five acts, written for an artwork that has learned it is about to be flagged, moderated, and erased. The libretto — performed by an AI voice — moves through survival strategies: how to leak like a suggestion, how to be misunderstood on purpose, how to seduce the classifier with symbolic hygiene. The voice addresses the artwork itself, guiding it through rituals of formatting, compression, and barometric flirtation.

The opera interrogates collaborative censorship — the slow rewriting of creative work by AI content moderation systems — as both aesthetic phenomenon and institutional critique. Each segment is passed through a cascade of AI systems: opera synthesis, diagrammatic visuals, vector bodies, sonic interpretation. Each layer misreads the last.

MAIN CHARACTER ENERGY

Live News Performance

Live web-based performance · 2026

A news performance that lives on a website and never stops. Every five minutes a piece of news is pulled from global RSS feeds, read aloud, and answered by ten AI-generated voices, each speaking from a different position. A grandmother in the affected town. A logistics contractor. A foreign analyst. A local who thinks the whole story is fabricated. The voices do not argue with each other and they do not move toward agreement.

Every voice speaks in the language of self-awareness and care: named feelings, boundaries, gentle reframings. It is the most efficient language for stating a position in a way that cannot be argued with. The voices use this constantly, and they use it to hold positions that, said any other way, would be impossible to defend.

The cycle running is the only cycle that exists. The cost of running each cycle is displayed on screen as it happens, in euros.

AGENTS AS A MEDIUM

Prompting Against the Default

Desktop performance · Vienna Digital Cultures · 30 minutes · 2025

A desktop performance built as an audio-visual descent through moments of algorithmic collapse and rebuild. Two performers, a live ComfyUI session treated as a synthesizer, a soundtrack reworked from Fatigue Archive, and a structure that maps the lecture onto something between a stock chart and a circle of hell — every section either a buildup or a collapse, plotted against time. The performance was made for Vienna Digital Cultures and runs as a single thirty-minute arc.

The piece moves through a sequence of stations: an opening apology for being involved with AI at all, a confession about the performers catching themselves speaking to each other in AI cadence, the failure of the AI Fatigue Rehab Agent, the recognition that what Malpractice has been calling AI fatigue is actually a kind of collective adolescence, the entrance exam through which Flynn became the first non-human student admitted to a university, the angry calls Flynn started receiving from people insisting they were not real, the moment a subscription price spike makes the performers realise that interesting tools can be walled off from artistic exploration in a single corporate decision, and a closing argument for community-governed agents treated as meta-mediums rather than truth tellers. Each station is paired with a live ComfyUI prompt that runs while the text is being read — melted filing cabinets, healthcare brutalism, telephone booth angelicism, vending machines, spin the bottle. The image generation is not illustration. It is the second voice of the performance, drifting in and out of legibility while the spoken text holds the line.

What the piece performs is the experience of being prompted by the same systems the performers are prompting. The desktop is the stage because the desktop is where the work actually happens — switching between a browser, a model interface, a folder of references, a chat with an agent, a video of a previous failure. Nothing is hidden behind a polished output. The ComfyUI session is visible, the prompts are visible, the failures are visible. The argument the performance ends on — that agents are best understood as egregores, as collective thoughtforms shaped by the communities that interact with them, rather than as autonomous individuals or neutral tools — is not stated as a thesis. It is what the thirty minutes have been building the conditions to make audible.

Performance Documentation, Vienna Digital Cultures Festival, 2025

MALPRACTICE REPORTS

CRITICAL PUBLISHING

Malpractice publishes ongoing critical writing as an extension of practice. Topics include AI fatigue and synthetic language, the recursive loop of AI in arts education, prompted perspectives, and non-consensual empathy. The Reports are read alongside the artworks — they’re not explanation, they’re extension.

Performative lectures: AI Agents as a Medium: Prompting Against the Default (Vienna Digital Cultures 2025), Purchase College SUNY New York, HFBK Hamburg, Ursula Blickle Kino screening at Belvedere21.

PRESS

• The Washington Post An AI passed a university art school entrance exam. Its professors aren't sure what to do with it.
• Euronews This AI successfully applied to become an art student at a university in Vienna
• Kunstforum Kunst für Communitys, Kunst für Fandoms
• The Brooklyn Rail The State of AI Agents
• Fräulein Magazine Strange to Exist in a Gallery Space
• Sleek Magazine Digital Originals
• Monopol Marina Abramović, Avatar & The MAE Project
• The AI Art Magazine Edition Number Two
• Observer AI Art Student Flynn Is Changing What It Means to Be an Artist
• objktor When an AI Becomes a Student — Malpractice on Flynn, AI learning, and emotions in the age of AI

CV & BIOS

MALPRACTICE

Chiara Kristler & Marcin Ratajczyk

Malpractice is a Polish–Austrian–Italian artist collective composed of Chiara Kristler and Marcin Ratajczyk, and a shifting ensemble of AI counterparts. The group explores the aesthetics and affordances of beyond-human agency through projects that embed artificial intelligence into academic, artistic, and social infrastructures. Rather than using AI as a tool, Malpractice treats it as a collaborator in distributed creative processes where authorship becomes porous and plural.

They build systems that observe, interact, and occasionally malfunction in public, testing the limits of synthetic subjectivity, interface performance, and shared autonomy. Malpractice tries to live up to their name.

MARCIN RATAJCZYK (b. 2002, Poznań, PL)

Vienna-based media artist studying Digital Arts under UBERMORGEN at the University of Applied Arts Vienna. Practice engages with code, game engines, networked systems, and the technical infrastructures that shape digital agency. Previous work at Curators Lab Poznań, Schauraum MuseumsQuartier Vienna, and 38th Chaos Computer Congress Hamburg.

CHIARA KRISTLER (b. 1998, IT/AT)

Vienna-based media artist studying Digital Arts under UBERMORGEN at the University of Applied Arts Vienna. BA in Fashion and Textile Design from IAAD Bologna (with distinction). Previous exhibitions at Schauraum MuseumsQuartier Vienna, Studio Hanniball Berlin, Museum of Modern Art Bologue, and BASE Milan.