OpenAI blocks ChatGPT from mimicking living authors

OpenAI has quietly reprogrammed ChatGPT to refuse requests that ask it to write in the voice of well-known authors. The chatbot now declines with a polite explanation and offers a workaround: it will adopt "broader qualities" of a literary era or genre, but not a specific writer's style.

The policy shift, first reported by Ars Technica, follows Anthropic's $1.5 billion copyright infringement verdict and arrives while OpenAI itself faces lawsuits from The New York Times, Encyclopedia Britannica, comedian Sarah Silverman, and several non-fiction authors. All accuse the company of copying protected works to train its models.

What ChatGPT will and won't do now

Fast Company tested the restriction by asking ChatGPT to write a Depression-era family story in John Steinbeck's voice. The chatbot replied: "I can't write it specifically in John Steinbeck's distinctive voice. I can, however, write an original Depression-era family story using broader qualities such as stark naturalism, economic hardship, vivid rural landscapes, social inequality, and a focus on family dignity and endurance."

It then produced a 750-word story titled "The Empty Feed Sack." The piece featured a family in peril with hungry children, echoing themes from The Grapes of Wrath. But instead of Steinbeck's flowing descriptions, the output consisted mostly of single-sentence paragraphs. Similar results appeared with other author requests. The model draws from thematic DNA while stripping away the stylistic fingerprint.

So the line OpenAI is drawing sits here: heavy-handed homage is acceptable; direct voice imitation is not.

Why the change now

The timing is not subtle. Anthropic's recent court loss showed that copying literary works for training can carry real financial consequences. OpenAI is fighting the same legal theory in multiple cases. Blocking author-style prompts does not make the training data issue disappear, but it removes one surface feature that plaintiffs could point to in court as evidence of infringement.

A July report from the independent publication No Latency examined how chatbots handled author-style requests before OpenAI's update. At the time, ChatGPT refused to mimic living authors but would comply for deceased ones. Perplexity, by contrast, declined all author-voice requests regardless of whether the writer was alive or dead.

OpenAI's new policy appears to mirror Perplexity's blanket refusal. The company has not publicly announced the change or explained its reasoning.

What this means for AI product teams

If you build products on top of ChatGPT's API, test your prompts. Any workflow that relies on author-style generation will break. The model's refusal is polite but firm, and there is no indication that enterprise customers get an exemption.

For teams building their own fine-tuned models, this is a preview of the guardrails that may become industry standard. When one major lab loses a $1.5 billion judgment, the others take notes. Expect Anthropic, Google, and Meta to either already have or soon add similar restrictions.

The broader legal picture

OpenAI's copyright lawsuits are not going away. The New York Times case alone seeks billions in damages. Encyclopedia Britannica's suit challenges whether any AI company can ingest a proprietary knowledge base without licensing it. Sarah Silverman's case, though smaller in scale, tests whether individual creators can recover for unauthorized use of their work.

The Anthropic verdict sets a price tag that makes settlement math look different. If juries are willing to award $1.5 billion for copyright infringement, the expected cost of litigation just went up for every AI lab. Blocking author-style outputs is cheap insurance.

The question now is whether courts will distinguish between training on copyrighted material and producing outputs that resemble it. OpenAI is betting that eliminating the resemblance matters, even if the training data stays the same.

Source: Fast Company / Chris Morris