The first free tool of its kind gives newsrooms, HR, insurance and AI teams, among others, a practical way to help eliminate biased language from written content and AI training data – in seconds
The Vector Institute today launched UnBias-Plus, a free, open-source AI tool built by Vector’s AI Safety research scientists that helps detect, explain, and rewrite biased language in written content and AI training datasets – in seconds.
The release comes at a decisive moment. Canada’s National AI Strategy: AI for All explicitly identifies addressing algorithmic bias as a pillar of responsible AI development, recognizing that unchecked bias in AI systems undermines equity, safety and public trust. The scale of the challenge is staggering: bias in AI training data has been measured at rates ranging from 3.4 per cent to 38.5 per cent, and even safety-tuned large language models (LLMs) exhibit implicit racial and gender bias – a troubling reality given that LLMs risk perpetuating and amplifying the very human biases embedded in the text they learn from.
UnBias-Plus is the first free tool of its kind available to both organizations and the general public. The tool analyzes written text for biased language across dimensions of race, gender, age and political framing, explains precisely why language has been flagged, suggests neutral alternatives, and returns a fully rewritten version of the original content. Rather than simply producing a revised copy, UnBias-Plus shows users why language may be biased, helping build a clearer understanding of how bias appears in everyday communication. The platform is available in two formats:
- A browser-based version for public use, supporting content of up to 750 words – practical for everyday writing and for organizations evaluating the tool before broader adoption; and
- A developer installer for integration into company applications, systems and workflows.
“What drove us to build this was simple: the people most harmed by biased language are often the last to know it’s there. A patient doesn’t see the assumptions buried in their clinical notes. A job candidate doesn’t know why a door keeps closing. We wanted to make the invisible visible – a free tool that surfaces bias, explains it and helps fix it. However, bias doesn’t just live in a document someone writes today. It lives in the data used to train AI models and, when it goes undetected there, it gets learned, replicated, and deployed at scale across entire organizations. UnBias-Plus gives developers and data scientists the ability to find and clean that bias before it becomes part of the model.”
UnBias-Plus is built to help professionals and sectors where language bias carries the highest real-world consequences:
Editorial and media – Editors can integrate UnBias-Plus directly into their workflows to help identify and correct bias before publication, supporting more accurate, inclusive reporting.
Human resources, EDI and workplace teams – The tool can be used to help screen job descriptions, performance reviews, promotion cases, and internal communications for gendered, racialized or otherwise loaded language that can unintentionally create barriers for employees and candidates.
Health care communications – UnBias-Plus helps review clinical notes, care assessments, and patient communications for stigmatizing or discriminatory framing that can influence care decisions, insurance coverage, and outcomes. For example, if a caregiver wrote “noncompliant patient refused medication” – that wording follows the patient into every future appointment. UnBias-Plus might flag such wording and suggest an alternative such as “patient declined medication” so the record stays factual, not judgmental.
Insurance – Teams can help screen benefits letters, eligibility notices, and coverage communications to ensure language reflects policy and fact, not unconscious assumptions about who deserves care.
AI developers and data scientists – UnBias-Plus helps screen annotations, prompts, and source text so sanitized data can be used to train or fine-tune AI models, identifying biased patterns at the data layer – before those patterns are learned, replicated, and scaled into deployed systems. AI teams can also run model outputs through the tool to audit for biased language before it reaches users or becomes embedded in downstream workflows.
“Canada’s AI leadership depends on building systems that create more equitable outcomes – and that means ensuring the research behind those systems is accessible to everyone. Making UnBias-Plus free, transparent and open-source is a deliberate expression of Vector’s mission as an independent, not-for-profit institute: research on how bias operates in AI and language affects all of society and the tools that come from it should be too. By opening this work to developers, organizations and the public, we’re ensuring the ability to build fairer AI belongs to everyone.”
Explore UnBias-Plus:
UnBias-Plus is Vector Institute’s first release in a broader suite of Safe AI tools. Updates will expand capabilities with audio and video bias detection and misinformation detection tools planned for future release. To access other available products, please visit our Open Source Software page.
UnBias-Plus currently supports English-language text only. The tool does not verify factual accuracy, detect misinformation or identify AI-generated content.
