Vector Institute releases UnBias-Plus, a free, open-source AI tool to detect and rewrite bias in text

2026-06-30 · Source: Vector Institute for Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Novice, quick

Summary

The Vector Institute has launched UnBias-Plus, a free, open-source AI tool designed to detect, explain, and rewrite biased language in written content and AI training datasets. This tool, the first of its kind available to both organizations and the public, addresses bias across dimensions of race, gender, age, and political framing. It provides neutral alternatives and a fully rewritten version of the original text, helping users understand how bias manifests. UnBias-Plus is offered as a browser-based version for public use, supporting up to 750 words, and a developer installer for integration into company applications. This release is timely, given that bias in AI training data has been measured between 3.4% and 38.5%, and even safety-tuned large language models exhibit implicit biases. It is the initial offering in Vector's broader suite of Safe AI tools.

Key takeaway

For AI developers and data scientists building or fine-tuning models, UnBias-Plus offers a critical pre-deployment check. You can integrate this free, open-source tool into your workflows to screen training data annotations, prompts, and model outputs for racial, gender, age, or political bias. This proactive step helps prevent the learning, replication, and scaling of human biases into your deployed AI systems, ensuring more equitable outcomes and maintaining public trust.

Key insights

UnBias-Plus offers a free, open-source AI solution to detect, explain, and rewrite language bias in text and AI training data, making invisible biases visible.

Principles

• Algorithmic bias undermines equity, safety, and public trust.
• Bias in AI training data scales into deployed systems.
• Tools should surface, explain, and help fix bias.

Method

UnBias-Plus analyzes text for bias across race, gender, age, and political framing, explains flags, suggests neutral alternatives, and provides a rewritten version. This helps users understand and correct bias.

In practice

• Screen job descriptions and performance reviews.
• Review clinical notes for stigmatizing language.
• Audit AI model outputs for biased language.

Topics

• AI Bias Detection
• Responsible AI
• Open-Source Tools
• Language Bias
• AI Training Data
• Algorithmic Fairness

Best for: NLP Engineer, CTO, VP of Engineering/Data, AI Engineer, Machine Learning Engineer, HR Professional

Related on AIssential

• UnBias-Plus: Detect, Explain, and Rewrite Bias — UnBias-Plus is an open-source toolkit that unifies multi-class bias detection, localization, explanation, and rewriting for natural language content.
• Bias in the machine (edited) — Bias in AI systems stems from historical human prejudices embedded in training data and model design, leading to real-world harm.
• A Systematic Analysis of Biases in Large Language Models — LLMs, despite alignment efforts, inherit and perpetuate diverse biases from human-generated training data.
• BAID: A Benchmark for Bias Assessment of AI Detectors — AI text detectors exhibit significant bias against underrepresented groups, necessitating systematic evaluation before deployment.
• BiasEdit: A Training-Free Bias-Detect-and-Edit Framework for Learning Fair Visual Classifiers — BiasEdit automatically detects and edits unknown dataset biases using visual-linguistic analysis and text-guided image editing, creating fair classifiers without manual annotation.
• Can Training Data for AI Ever Be Without Bias? — Bias-free AI training data is unattainable; focus on understanding and managing inherent biases.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Vector Institute for Artificial Intelligence.