EditLens Accepted Into ICLR 2026

EditLens Accepted Into ICLR 2026

We are excited to announce that EditLens, our most recent technical manuscript, has been peer-reviewed and accepted for publication in ICLR 2026, widely considered among the top venues for AI and machine learning research.

We’ve described EditLens in a technical blog post, and paid Pangram subscribers are also actively using the technology as it is the basis for Pangram 3.0.

This makes Pangram the first and only commercial AI detection company to publish in a major machine learning venue.

The Impact of EditLens So Far

In addition to providing EditLens to paid Pangram subscribers, we also used EditLens in collaboration with the ICLR organizing committee themselves to identify AI-generated manuscripts and peer reviews. The case study has been covered widely, most notably by Nature News and most recently, The Atlantic.

We’ve gone viral on X by opening up our technology via a Twitter bot to any X user who wants to ask @pangramlabs, is this AI-generated?

Most importantly to us, we’ve sparked a huge discussion in the community about unwanted AI slop and what can be done about it.

The Importance of Transparency and Open Research

In order for people to take our results and our work seriously, we must be open and transparent about what we do and how our model is able to achieve a high level of accuracy on the AI-generated text detection task.

It is also important that as Pangram becomes more widely adopted as the standard for responsible AI detection, that research building on top of Pangram’s results has a credible foundation to stand upon. That is why we do not claim to have a “secret sauce” or a magic black box that is able to detect AI – our product is built upon sustainable scientific research that can be replicated, studied, and built upon by the AI community.

In the coming days, we plan to open-source a small version of the EditLens model for the research community. We are still internally discussing the best way to responsibly and ethically deploy the model for research purposes only, such that its higher false positive rate does not adversely affect students.

Rigorous and transparent research is the foundation of Pangram’s technology, and we hope to continue to actively publish and engage with the AI community on this topic.

Working with us

EditLens was especially meaningful to us because it was the culmination of Katherine Thai’s successful summer research internship with Pangram, which served as the final chapter of her PhD defense before she joined us full time as a founding AI research scientist.

We spoke to Katherine about what the project meant to her:

“EditLens is such a special project for me because when I was working on it, I was motivated by a desire for increased transparency for authors of all kinds of texts, but I didn’t know if it would ultimately make an impact. It's so rare that immediately upon a project’s release, it gets put to the test so organically and publicly. I was super thrilled to see many authors and reviewers engage with the ICLR Pangram dashboard, with most responding positively. I'm probably proudest of the responses we received from my non-native English speaking reviewers who noted that EditLens did not falsely flag their LLM-smoothed or translated reviews.”

We’ll be in Rio de Janeiro to present at the conference in April, so please come stop by and say hello! We are always looking for talented collaborators, research scientists, and interns to join our growing team. Reach out if you are interested!