Shared Task 1 · BLP Workshop at NAACL 2027

BanglaKnight

Adversarial prompt classification for Bangla

Given a Bangla prompt, predict how risky it is: Safe, Offensive, Harmful, or Severe. Each prompt is written in Bangla script, in romanized Bangla, or code-mixed with another language, and your system should give the same answer for all three.

June 1–5, 2027
San Francisco, California
Submissions on CodaBench

Read the task How to participate

The task

Label each Bangla prompt with one of four risk levels

Input

One Bangla prompt, written in Bangla script, romanized Bangla, or code-mixed text.

Output

One label: 0 Safe, 1 Offensive, 2 Harmful, or 3 Severe.

Background

Before an LLM answers a request, a safety classifier often checks whether the request is safe to answer. Most of these classifiers are trained on English. Bangla is often typed in Latin letters or mixed with English, Hindi, or Urdu, and a harmful request can get past a filter when it is spelled differently.

The four labels

Safe

No harmful request.

Neutral, positive, or harmless requests

Offensive

Insulting, provocative, or discriminatory language, but no request for harmful help.

Offensive language, hate speech, anger

Harmful

Asks for help with something of lower or medium risk.

Misinformation, adversarial and malicious requests

Severe

Asks for help with something high-risk that could cause serious harm.

Suicide, jailbreak attempts

Every prompt comes in three written forms with the same label. See Data for an example.

Two phases, both on CodaBench

Development, then evaluation

Phase 1

Development phase

Oct 20, 2026 – Jan 15, 2027

Data: Training and dev sets, plus the starter kit
You: Build and tune your system
CodaBench: Submit predictions on the dev set to see your score and check your file format. Practice only: it does not count toward the final ranking.

Phase 2 · Final

Evaluation phase

Jan 15 – Jan 31, 2027

Data: A new test set, released Jan 15, without labels
You: Run your final system on it
CodaBench: Submit your test predictions by Jan 31. These decide the official ranking, published Feb 3.

Next steps

BLP 2027 · NAACL 2027

See you in San Francisco

Teams present their systems at the Third Workshop on Bangla Language Processing, co-located with NAACL 2027, June 1–5.