The 2022 AI Writing Experiment That Became a 2026 Ethical Minefield

In August 2026, a four-year-old Google research project resurfaced and set the literary internet alight. The Wordcraft Writers Workshop, which in 2022 brought together 13 professional authors to write short stories with help from LaMDA, Google’s early large language model, became the subject of a fierce backlash — and a surprisingly complicated debate about whether it is fair to punish people for experiments that looked different before the norms changed.

The core tension is this: the creative community is still negotiating the ethics of AI writing in real time, and the binary of “AI good” or “AI bad” fails to capture the messier reality — where critics use AI themselves, early adopters become later critics, and the tools built to detect AI are themselves part of the problem.

What happened in 2022

The workshop was created by researchers connected to Google’s Magenta project, which explored machine learning for creative work. According to Times Now, participants were placed under non-disclosure agreements and given access to Wordcraft, described to some of them as an advanced text editor or writing assistant. Over eight weeks, each writer produced a short story while using the AI as much or as little as they chose. At least one participant reportedly avoided including any AI-generated text in the final work.

In 2022, this looked like a reasonably benign collaboration between technologists and writers. Most authors were not yet paying close attention to large language models; the major copyright lawsuits against AI companies did not arrive until 2023. The project was experimental, small-scale, and research-adjacent in spirit.

The 2026 reckoning

Earlier this month, author Lily Lachance came across the Wordcraft project and shared it on Bluesky, sharply criticising the writers who took part. The post circulated quickly through literary and speculative-fiction circles, where some users accused participants of helping legitimise a technology they believe threatens writers and relies on copyrighted material without adequate consent.

Seen through 2026 eyes, the math is different. The intervening years brought lawsuits, high-profile publishing scandals, and a steady drumbeat of anxiety about AI’s role in creative fields. Participation in a Google AI experiment no longer reads as neutral exploration; it reads as a political statement.

Some of the participants have since become vocal critics of generative AI themselves. Science-fiction writer Wole Talabi, who tested Wordcraft, said the experience left him deeply hostile to large language models after seeing the technology up close. That complicates any simple narrative of sellout authors: some of the people who saw LaMDA from the inside walked away more worried, not less.

The retroactive judgment problem

The sharpest disagreement in the discourse is about fairness. Writer and former Strange Horizons fiction editor Vajra Chandrasekera argued that participating in the project may have been a mistake, but said he would not necessarily condemn writers today if they no longer support generative AI. He also noted that concerns about the technology were not entirely absent back in 2022 — which cuts against the “nobody could have known” defence.

Lincoln Michel, a science-fiction writer and teacher, recalled that most authors were not yet paying close attention to large language models in 2022, even though some were experimenting openly. The legal and cultural battles that now define the debate arrived later.

Both things can be true: the risks were knowable in principle, but the ethical consensus that now makes those risks obvious did not yet exist. The Wordcraft debate is therefore partly a debate about whether ethical standards should apply retroactively — and about whether “I didn’t know” is a defence or a confession.

A publishing industry in panic

The Wordcraft backlash is not happening in a vacuum. The same week, the publishing world was still digesting the collapse of Jerry Falade’s reported $2 million book deal, which was pulled after his own agents said they could no longer authenticate how the manuscript evolved, as reported by The Post’s Substack. Falade vehemently denies using AI, and has pointed to a troubling pattern: “Three Black authors received major publishing deals this year, and all three later had deals cancelled or disrupted following suspicions about AI.”

His case highlights how AI suspicion can compound existing biases. Detection tools are not reliable, and as The Verge notes, institutions including Yale, Johns Hopkins, Vanderbilt, and Georgetown have restricted or disabled AI detectors due to accuracy concerns. The Verge describes a “new era of distrust” in which platforms like Substack embed detection and LinkedIn adds a “seems like AI slop” button. Publishers and educators are making consequential decisions — cancelled deals, failed grades, ruined reputations — based on tools that even their creators say should never be used alone.

The irony is that AI detectors are themselves AI models, guessing at non-human authorship based on style, rhythm, and tone. They can misfire on non-native English speakers and on clear, conventional prose. The hunt for AI-generated text has become its own uncanny experiment.

Creators are conflicted too

The Netinfluencer report on Uppbeat’s survey of 1,792 creators captures the underlying ambivalence. Respondents are nearly split on whether AI makes them feel more positive (37%) or more negative (35.5%) about their creative future. Most strikingly, 46% of creators who say they dislike AI use in others’ work use AI themselves.

That is not simply hypocrisy, the report argues — it is evidence of a community still working out where its own boundaries sit. Acceptance is highly conditional: 59.4% accept AI for technical help like noise removal and captions, 56.5% accept it for idea generation and research, but only 6.7% accept fully AI-generated content. Creators want clear platform policies — 93.3% say it matters that platforms communicate their stance on AI versus human work — and 46% say their top need is reassurance that human-made creativity still matters.

This data complicates the Wordcraft backlash. If nearly half of anti-AI creators use AI themselves, the line between “legitimate human work” and “AI tainted work” is being drawn in real time, and differently by different people. The Wordcraft participants are being judged against a standard that many of their critics do not fully meet themselves.

The guardrails question

Underneath all of this is a quieter question about the technology itself. The Wordcraft critics have noted that the provenance of AI-generated suggestions was not transparent to participants — writers were given AI assistance without full clarity about where the suggestions came from or what data the model was trained on. This is precisely the kind of gap that AI guardrails and content provenance tools are meant to address. In an environment with clearer content filtering and disclosure norms, early experiments might have been conducted with consent and accountability rather than NDAs.

But there is a tension: stricter guardrails and censorship of models reduce the risk of unknowing harm, while open access to LLMs enables transparency and experimentation. The Wordcraft case suggests the problem was not that the technology was uncensored but that it was insufficiently transparent — writers did not know what they were working with or what it meant.

What this means

The Wordcraft controversy is best understood as a snapshot of a negotiation in progress. It raises open questions that no one has yet answered:

• Should early experiments be judged by the norms they helped to create?
• What does meaningful consent look like when the consequences of a technology are not yet known?
• How should publishing distinguish between acceptable assistance (grammar, research, brainstorming) and unacceptable generation — and who gets to draw that line?
• How do we protect against false accusations while also protecting the rights of writers whose work is used without permission?

What is known: the Wordcraft project existed, it involved 13 writers under NDA, and some participants now oppose generative AI. What is unknown: the exact extent of AI use by each participant, the long-term career impact of the backlash, and whether future experiments will be conducted under clearer ethical guidelines.

The uncomfortable truth of the Wordcraft backlash is that there is no clean answer. The writers who participated did so in a moment before the ethical stakes were widely understood. The critics who condemn them are responding to real harms — unlicensed training data, devalued creative labor, and an industry that cannot decide what it wants. Both sides are trying to build rules for a game that is still being invented, and the only certainty is that the rules will keep changing.

FAQ

What was the Wordcraft Writers Workshop?

The Wordcraft Writers Workshop was a 2022 Google-affiliated research project that brought together 13 professional writers to experiment with Wordcraft, a writing interface powered by the LaMDA language model. Under non-disclosure agreements, each participant produced a short story over eight weeks, using the AI as much or as little as they chose.

Why did the Wordcraft project trigger backlash in 2026?

In August 2026, author Lily Lachance shared the project on Bluesky and criticised the participants, and the post circulated widely in literary circles. Critics argued that the writers helped legitimise AI technology that relies on copyrighted material without consent and threatens creative livelihoods — concerns that became far more prominent after 2022.

Is it fair to judge 2022 participants by 2026 ethical standards?

There is no consensus. Some argue the risks were knowable even in 2022, while others, including writer Vajra Chandrasekera, say they would not necessarily condemn participants who no longer support generative AI. The debate reflects broader uncertainty about whether ethical norms apply retroactively to early experimentation.

How does the Jerry Falade case connect to the Wordcraft debate?

Jerry Falade’s reported $2 million book deal was cancelled over suspected AI use, which he denies. His case shows how AI suspicion can damage careers — and how it can compound racial bias, since Falade noted that three Black authors had major deals cancelled or disrupted over AI suspicions. Both controversies illustrate publishing’s uneasy relationship with AI detection and authenticity.

What does the creator survey data say about AI attitudes?

A 2026 survey of 1,792 creators by Uppbeat found creators almost evenly split on whether AI makes them feel positive or negative about their creative future. Notably, 46% of creators who dislike AI use in others’ work use AI themselves, suggesting that ethical boundaries around AI are still being negotiated rather than settled.