Why Major Conferences Are Banning ChatGPT-Generated Papers—and What It Means for AI Research

Several institutions, including major international AI and machine‑learning conferences as well as educational bodies, have announced bans on content generated by ChatGPT and other large language models (LLMs).

This year, multiple top‑tier ML conferences will prohibit submissions that contain text produced by LLMs such as ChatGPT, citing risks of plagiarism and the need to preserve academic integrity.

Since OpenAI released GPT‑3 in 2020, text‑generation models have rapidly improved and become widely adopted, leading to a proliferation of products that can adapt to various writing styles. Some users have already employed these tools to generate code or draft papers, prompting organizations like ICML to explicitly forbid AI‑written manuscripts.

"Papers containing text generated by large‑scale language models (LLM) such as ChatGPT are prohibited unless the generated text is presented as part of experimental analysis," the ICML chair stated in an official announcement.

Nevertheless, many scholars still use AI to polish their work, feeding their drafts into models for stylistic or grammatical improvements. ICML participants acknowledge that the ban may evolve as the technology and its implications become clearer.

In New York State, public‑school networks have been ordered to block access to ChatGPT, with officials warning that the tool could undermine critical thinking and problem‑solving skills essential for lifelong achievement.

The Emerging "ChatGPT Ban Storm"

LLMs are trained on massive web corpora, learning patterns to predict subsequent text. While there is no definitive evidence that they directly copy large passages, concerns persist about potential copyright infringement and the originality of AI‑generated output.

Questions arise regarding ownership of AI‑produced text and images: who holds the rights—the user, the model developer, or the creators of the training data?

ICML officials emphasize that the lack of time to study the impact of LLMs on peer review forced a precautionary ban on AI‑generated manuscript text.

Currently, no reliable tool can automatically detect AI‑written text; reviewers must manually flag suspicious submissions. AI‑generated drafts often contain factual errors and require substantial human editing.

Detection Efforts: GPTZero

Edward Tian, a Princeton computer‑science student, created GPTZero , an application that quickly identifies whether a document is human‑written or produced by ChatGPT.

Tian reported that GPTZero correctly classified a New Yorker article by John McPhee as human‑written, while a LinkedIn post was identified as AI‑generated. After an initial surge of traffic that overloaded the service, Tian updated the model to reduce false‑positive rates and improve output quality.

GPTZero evaluates text using two metrics: perplexity (a measure of randomness in sentence construction) and burstiness (the variation in sentence patterns). Human writing typically exhibits higher burstiness, reflecting diverse phrasing.

Broader Responses

OpenAI announced plans to embed watermarks in ChatGPT outputs to combat plagiarism. Meanwhile, platforms like Stack Overflow have banned AI‑generated answers due to frequent factual inaccuracies.

These developments illustrate a rapidly evolving landscape where the academic community, educators, and developers grapple with balancing AI’s capabilities against the need for originality, accountability, and rigorous scholarship.