# AI Fairness 360 (AIF360)

> Source: https://aiwiki.ai/wiki/aif360
> Updated: 2026-06-27
> Fact-checked: 2026-11-05
> Categories: AI Ethics, AI Tools & Products, Open Source AI
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "AI Fairness 360 (AIF360)." aiwiki.ai, 27 Jun 2026. https://aiwiki.ai/wiki/aif360
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**AI Fairness 360**, abbreviated **AIF360**, is an open-source Python and R toolkit, originally created by IBM Research and released in September 2018, that detects and mitigates unwanted bias in [machine learning](https://aiwiki.ai/wiki/machine_learning) models and the datasets they are trained on [1][5]. Its initial release shipped with more than 70 fairness metrics and nine bias-mitigation algorithms spanning three intervention stages (pre-processing, in-processing, and [post-processing](https://aiwiki.ai/wiki/post-processing)), all wrapped in a single uniform API [1][5]. In June 2020, [IBM](https://aiwiki.ai/wiki/ibm_ai) donated the project to the [Linux Foundation](https://aiwiki.ai/wiki/linux_foundation) AI & Data Foundation (LF AI & Data), and it is now community-governed under the Trusted-AI [GitHub](https://aiwiki.ai/wiki/github) organization at `Trusted-AI/AIF360`, where the project README now lists 15 supported bias mitigation algorithms [3].

AIF360 was one of the first widely adopted attempts to consolidate the academic literature on [algorithmic fairness](https://aiwiki.ai/wiki/algorithmic_fairness) into a single library with a uniform API, and it has since become a standard reference implementation for both researchers comparing methods and practitioners auditing production systems. IBM Research described the package on launch as "a comprehensive open-source toolkit of metrics to check for unwanted bias in datasets and [machine learning models](https://aiwiki.ai/wiki/model), and state-of-the-art algorithms to mitigate such bias" [5]. The project sits alongside sister toolkits such as AI Explainability 360 (AIX360) and the Adversarial Robustness Toolbox (ART), which were donated to LF AI & Data at the same time [29].

The toolkit was introduced in the paper "AI Fairness 360: An Extensible Toolkit for Detecting, Understanding, and Mitigating Unwanted Algorithmic Bias" by Rachel K. E. Bellamy and 17 co-authors (Bellamy et al., [arXiv](https://aiwiki.ai/wiki/arxiv):1810.01943, submitted 3 October 2018), later expanded for the IBM Journal of Research and Development in 2019 [1][2]. The library has grown steadily since, with regular minor releases adding datasets, intersectional metrics, an R interface, and tighter [scikit-learn](https://scikit-learn.org) compatibility.

## What is AI Fairness 360 used for?

AIF360 is used to measure and reduce algorithmic [bias](https://aiwiki.ai/wiki/bias) at three points in a typical machine-learning workflow: in the training data, inside the learning algorithm, and in the predictions of an already-trained model. A data scientist uses the toolkit to compute fairness metrics (for example, whether a model approves loans at a lower rate for one protected group than another), and then applies one of its mitigation algorithms to bring those metrics closer to parity while monitoring the impact on accuracy [1]. The stated goal of the project, per the Bellamy et al. paper, is "to help facilitate the transition of fairness research algorithms to use in an industrial setting and to provide a common framework for fairness researchers to share and evaluate algorithms" [1].

The toolkit is released under the Apache 2.0 license and is installable with `pip install aif360` in Python [3]; the R package was published on CRAN in 2020 but archived there on 20 June 2023, so R users now install it from the CRAN archive or from GitHub [32]. Typical applications include credit scoring, hiring, recidivism prediction, healthcare resource allocation, and any other high-stakes domain where a protected attribute such as race, sex, or age could influence model decisions.

## History: when was AIF360 released?

The origins of AIF360 trace to a 2018 collaboration between IBM India Research Lab and the IBM T.J. Watson Research Center. Concerns about bias in high-stakes machine-learning applications such as credit scoring, hiring, and recidivism prediction had moved from academic conferences (FAccT, then known as FAT*) into mainstream press, especially after ProPublica's May 2016 investigation into the [COMPAS](https://aiwiki.ai/wiki/compas) recidivism risk score [30]. IBM's response was to package the most influential fairness papers from the previous decade into a single Python library that any data scientist could `pip install`.

The public launch happened on 19 September 2018 with a blog post on the IBM Research site and the release of the GitHub repository (originally `IBM/AIF360`), followed on 3 October 2018 by the arXiv preprint by Bellamy et al. [5][1]. The launch included an interactive web demo, three end-to-end Jupyter tutorials covering credit scoring, medical-expenditure prediction, and facial gender classification, and an extensive documentation site at `aif360.readthedocs.io` [4]. IBM Research framed the release as a community effort, writing: "we hope that the package is not only a way to bring all of us researchers together, but also a way to translate our collective research results to data scientists, data engineers, and developers" [5].

In June 2020 IBM contributed AIF360 to the Linux Foundation AI & Data Foundation as an incubation project, alongside AIX360 (AI Explainability 360) and ART (Adversarial Robustness Toolbox) [29]. The LF AI Technical Advisory Committee formally approved hosting the three Trusted AI toolkits in June 2020 [31], and LF AI announced them as incubation projects on 22 September 2020 [6]. The GitHub repository was subsequently moved from `IBM/AIF360` to `Trusted-AI/AIF360`, signalling that the project was no longer governed solely by IBM. Later versions continued to add fairness algorithms from the wider research community: Celis et al.'s meta fair classifier arrived in version 0.2.0 (January 2019), before the donation, and Agarwal et al.'s reductions approach in version 0.4.0 (March 2021) [16][17][35]. An R interface was added in 2020, version 0.5.0 (September 2022) added FairAdapt and MDSS bias subset scanning, and version 0.6.0 (February 2024) added the SenSeI/SenSR algorithms and deterministic [re-ranking](https://aiwiki.ai/wiki/re-ranking) for fair ranking [35].

| Year | Milestone |
|------|-----------|
| 2018 (Sept-Oct) | Open-sourced by IBM Research in September; [arXiv](https://aiwiki.ai/wiki/arxiv) preprint 1810.01943 posted 3 October; over 70 metrics and 9 algorithms in the first release [1][5] |
| 2019 | Expanded paper appears in *IBM Journal of Research and Development* vol. 63, no. 4/5 [2] |
| 2020 (May-June) | R package added to the repository and published on CRAN (archived by CRAN in June 2023) [32] |
| 2020 (June) | Project contributed to LF AI & Data; repo moved to `Trusted-AI/AIF360` [6][29] |
| 2021 (March) | Version 0.4.0 adds Exponentiated Gradient and Grid Search reductions (ported from Fairlearn) and the Law School GPA dataset [35] |
| 2022-2024 | Version 0.5.0 (September 2022) adds FairAdapt and MDSS bias scanning; 0.6.0 (February 2024) adds SenSeI/SenSR and deterministic re-ranking; 0.6.1 (April 2024), the latest release as of November 2026, adds the FACTS detector [35] |

## What are the design goals of AIF360?

The Bellamy et al. paper states three goals for the toolkit [1]: to promote a deeper understanding of fairness metrics and mitigation techniques; to enable an open common platform for fairness researchers and industry practitioners to share and benchmark their algorithms; and to help facilitate the transition of fairness research algorithms to use in an industrial setting. It also describes the architecture as designed for ease of use and extensibility, and the launch included an interactive web experience intended as a gentle introduction to the concepts for line-of-business users, alongside tutorials for data scientists working with [scikit-learn](https://aiwiki.ai/wiki/scikit_learn)-style tools.

A recurring design choice across the codebase is the preference for fit/transform/predict semantics borrowed from scikit-learn. Mitigation algorithms expose `fit(dataset)` and `transform(dataset)` for pre- and post-processing methods, and `fit(dataset)` and `predict(dataset)` for in-processing methods. This makes it straightforward to drop AIF360 components into existing modelling pipelines without rewriting them.

## How is AIF360 structured?

AIF360 is organised around a small set of abstractions that the rest of the library composes on top of.

### Datasets

At the bottom of the stack is `StructuredDataset`, which wraps a pandas [DataFrame](https://aiwiki.ai/wiki/dataframe) together with metadata describing protected attributes (such as race, sex, or age), favourable and unfavourable labels, and privileged and unprivileged groups. `BinaryLabelDataset` is the most common subclass and is used for binary-classification fairness, which covers most of the published literature. `RegressionDataset` exists for continuous targets but supports a smaller set of algorithms. The dataset object is the unit on which both metrics and mitigation algorithms operate, which avoids the awkward situation of metrics taking raw arrays without knowing which column is the protected attribute.

### Metrics

Three metric classes cover most use cases. `BinaryLabelDatasetMetric` evaluates fairness properties of a single labelled dataset, for instance whether the base rate of the positive label differs across protected groups. `ClassificationMetric` takes two `BinaryLabelDataset` objects, one with ground truth and one with model predictions, and computes metrics like equalised odds, equal opportunity, and [predictive parity](https://aiwiki.ai/wiki/predictive_parity). `SampleDistortionMetric` measures how much a pre-processing transformation has altered individual records, which matters because many [fairness metric](https://aiwiki.ai/wiki/fairness_metric) interventions trade off group fairness against individual fairness.

### Algorithms

Mitigation algorithms are split into three subpackages corresponding to where they intervene in the pipeline.

- `aif360.algorithms.preprocessing` modifies the training data before any model sees it.
- `aif360.algorithms.inprocessing` modifies the learning algorithm itself, usually by adding a fairness term to the [loss function](https://aiwiki.ai/wiki/loss_function) or by training a classifier and adversary jointly.
- `aif360.algorithms.postprocessing` modifies the predictions of an already-trained classifier.

The Bellamy et al. paper attributes this taxonomy to d'Alessandro et al. (2017) [1].

### Explainers

Each metric can be wrapped in an explainer that produces human-readable text describing what the metric value means and which subgroup is disadvantaged. The toolkit also exposes JSON-formatted explanations for integration with downstream dashboards.

### scikit-learn compatibility layer

The `aif360.sklearn` namespace, introduced as an experimental API in version 0.3.0 (June 2020) [35], provides scikit-learn-compatible classes that operate directly on pandas DataFrames with `MultiIndex` rows where the protected attribute is part of the index. This makes it easier to chain AIF360 components with `sklearn.pipeline.Pipeline` and `GridSearchCV` for [hyperparameter tuning](https://aiwiki.ai/wiki/hyperparameter_tuning).

## What bias metrics does AIF360 provide?

The initial 2018 release exposed more than 70 fairness metrics, many of which are simple counting differences computed from the [confusion matrix](https://aiwiki.ai/wiki/confusion_matrix) sliced by protected group [1]. The most widely used ones are summarised below.

| Metric | What it measures | Origin |
|--------|------------------|--------|
| [Disparate impact](https://aiwiki.ai/wiki/disparate_impact) ratio | Ratio of positive-prediction rate for unprivileged group over privileged group; below 0.8 fails the four-fifths rule | US EEOC Uniform Guidelines on Employee Selection Procedures, 1978 [25] |
| Statistical parity difference | Difference in positive-prediction rates between groups | Calders & Verwer 2010; Dwork et al. 2012 [7][24] |
| Equal opportunity difference | Difference in true-positive rates between groups | Hardt, Price, Srebro 2016 [8] |
| Average odds difference | Mean of TPR and FPR differences between groups | Hardt, Price, Srebro 2016 [8] |
| Equalised odds | Joint constraint of equal TPR and equal FPR | Hardt, Price, Srebro 2016 [8] |
| Theil index | Entropy-based measure of inequality across individual benefits | Theil 1967; adapted by Speicher et al. 2018 [20] |
| Generalised entropy index | Family that includes Theil index as a special case | Speicher et al. 2018 [20] |
| Consistency | How similarly the model treats nearest-neighbour individuals | Zemel et al. 2013 [15] |
| Differential fairness | Smoothed empirical bound across all subgroups defined by protected attributes | Foulds et al. 2020 [21] |
| Smoothed empirical differential fairness | Bayesian-smoothed version of differential fairness | Foulds et al. 2020 [21] |
| Between-group / within-group generalised entropy | Decomposition of total inequality | Speicher et al. 2018 [20] |

The [disparate impact](https://aiwiki.ai/wiki/disparate_impact) metric deserves a note. The four-fifths rule is a US-specific legal heuristic adopted by the Equal Employment Opportunity Commission in 1978 to flag employment selection procedures with adverse impact on protected classes [25]. AIF360 ships it as a metric because it is widely cited; the rule has no legal force outside the United States, and even within the US the Uniform Guidelines treat it as a rule of thumb for flagging adverse impact rather than a binary pass/fail test [25].

## How does AIF360 mitigate bias?

The combination of pre-, in-, and post-processing algorithms is the part of AIF360 that has the largest effect on practical workflows. The first release contained nine such algorithms; the README of the `Trusted-AI/AIF360` repository now lists 15, each contributed by the broader algorithmic-fairness research community [5][3]. The following table lists the classes in the main `aif360.algorithms` module, grouped by category; the scikit-learn-compatible `aif360.sklearn` namespace adds further methods such as Fair Data Adaptation (FairAdapt) and SenSeI/SenSR [3][4]. Each algorithm is a faithful implementation of a published method, with the original paper as the canonical reference.

| Category | Algorithm | Authors and year | One-line description |
|----------|-----------|------------------|----------------------|
| Pre-processing | Reweighing | Kamiran & Calders 2012 [12] | Assigns weights to (group, label) cells so the joint distribution becomes statistically independent of the protected attribute |
| Pre-processing | [Disparate impact](https://aiwiki.ai/wiki/disparate_impact) remover | Feldman, Friedler, Moeller, Scheidegger, Venkatasubramanian 2015 [14] | Edits feature values to remove disparate impact while preserving within-group rank ordering |
| Pre-processing | Learning fair representations (LFR) | Zemel, Wu, Swersky, Pitassi, Dwork 2013 [15] | Learns a latent representation that obfuscates the protected attribute while preserving utility |
| Pre-processing | Optimised pre-processing | Calmon, Wei, Vinzamuri, Ramamurthy, Varshney 2017 [13] | Convex optimisation that jointly bounds discrimination, individual distortion, and utility loss |
| In-processing | Adversarial debiasing | Zhang, Lemoine, Mitchell 2018 [11] | Trains a classifier and an adversary jointly so the adversary cannot recover the protected attribute from predictions |
| In-processing | Prejudice remover | Kamishima, Akaho, Asoh, Sakuma 2012 [10] | Adds a prejudice index regulariser to [logistic regression](https://aiwiki.ai/wiki/logistic_regression) |
| In-processing | Meta fair classifier | Celis, Huang, Keswani, Vishnoi 2019 [16] | Takes a [fairness metric](https://aiwiki.ai/wiki/fairness_metric) as input and returns a classifier optimal for that metric |
| In-processing | Exponentiated gradient reduction | Agarwal, Beygelzimer, Dudik, Langford, Wallach 2018 [17] | Reduces fair classification to a sequence of cost-sensitive classification problems |
| In-processing | Grid search reduction | Agarwal, Dudik, Wu 2019 | Grid-search variant of the reductions approach for [binary classification](https://aiwiki.ai/wiki/binary_classification) and regression |
| In-processing | GerryFair classifier | Kearns, Neel, Roth, Wu 2018 [19] | Audits and mitigates fairness gerrymandering across rich subgroups |
| In-processing | ART classifier wrapper | IBM Research 2018 | Wraps an Adversarial Robustness Toolbox classifier so it can be measured by AIF360 |
| Post-processing | Equalised odds post-processing | Hardt, Price, Srebro 2016 [8] | Solves a linear program to randomise predictions until TPR and FPR match across groups |
| Post-processing | Calibrated equalised odds | Pleiss, Raghavan, Wu, Kleinberg, Weinberger 2017 [9] | Relaxes equalised odds to preserve calibration on classifier scores |
| Post-processing | Reject option classification | Kamiran, Karim, Zhang 2012 [18] | Assigns favourable outcomes to the unprivileged group and unfavourable outcomes to the privileged group inside a confidence band around the [decision boundary](https://aiwiki.ai/wiki/decision_boundary) |
| Post-processing | Deterministic re-ranking | Geyik, Ambler, Kenthapadi 2019 [4] | Rebalances ranked candidate lists to satisfy a [fairness constraint](https://aiwiki.ai/wiki/fairness_constraint) at every prefix |

Three of these names appear among the example algorithms IBM Research highlighted at launch: "Optimized Pre-Processing for Discrimination Prevention," "Certifying and Removing Disparate Impact," and "Learning Fair Representations" [5]. Most in-processing methods have specific dependencies. Adversarial debiasing requires [TensorFlow](https://aiwiki.ai/wiki/tensorflow) because AIF360's implementation is written in it, and optimised pre-processing requires CVXPY because it solves a convex program. The library ships these as optional extras (for example `pip install 'aif360[AdversarialDebiasing]'` or `pip install 'aif360[all]'`), so users who do not need those algorithms do not have to install TensorFlow [3].

## Which datasets does AIF360 include?

AIF360 ships several benchmark datasets that the fairness literature has converged on, mostly because they are public and have natural protected attributes. These include the following.

| Dataset | Domain | Protected attributes | Notes |
|---------|--------|----------------------|-------|
| Adult (Census Income) | US Census 1994 | Sex, race | Predicting whether income exceeds 50,000 USD; the most cited fairness benchmark |
| German Credit | Banking | Sex, age | UCI repository; predicting credit risk |
| COMPAS | Criminal justice | Race, sex | The dataset behind the ProPublica 2016 investigation; recidivism risk in Broward County, FL [30] |
| Bank Marketing | Marketing | Age | Portuguese banking institution telemarketing dataset |
| MEPS (19, 20, 21) | Healthcare | Race | Medical Expenditure Panel Survey; predicting high utilisation of care |
| Law School GPA | Education | Race, sex | Wightman 1998 LSAC; first-year GPA; supports regression fairness |

The datasets ship with helper code to download the raw files (because most cannot be redistributed under permissive licences) and to encode protected attributes consistently with the fairness literature.

## Is AIF360 available in Python and R?

Yes. AIF360 began as a Python package and added an R interface in 2020, and the project README states that "AI Fairness 360 package is available in both Python and R" [3]. The Python package is installed with `pip install aif360`. The R package was published on CRAN, but CRAN archived it on 20 June 2023, so it must now be installed from the CRAN archive or from GitHub with `devtools::install_github("Trusted-AI/AIF360/aif360/aif360-r")` [32]. Both expose the same core abstractions (datasets, metrics, and the three families of mitigation algorithms), although the Python package receives new algorithms first and remains the more actively developed of the two.

## How does AIF360 integrate with other tools?

Because AIF360's mitigation algorithms expose either scikit-learn or transform-style APIs, integration with the broader Python ML ecosystem is straightforward. A few specific integrations are worth calling out.

- **scikit-learn**: most in-processing algorithms can wrap arbitrary scikit-learn classifiers, and the `aif360.sklearn` module is designed to plug into scikit-learn pipelines.
- **TensorFlow / [PyTorch](https://aiwiki.ai/wiki/pytorch)**: adversarial debiasing has a TensorFlow implementation; community contributions have added PyTorch wrappers for several algorithms.
- **Fairlearn**: [Microsoft](https://aiwiki.ai/wiki/microsoft)'s Fairlearn project (open-sourced 2018, MIT licence) covers similar ground. The two libraries are not mutually exclusive: AIF360's two reductions algorithms (exponentiated gradient and grid search) were ported from Fairlearn in version 0.4.0 [35].
- **Aequitas**: the University of Chicago Center for [Data Science](https://aiwiki.ai/wiki/data_science) and Public Policy bias audit toolkit. Aequitas is more focused on producing audit reports for non-technical stakeholders, while AIF360 is more focused on mitigation.
- **AIX360** (AI Explainability 360): IBM's sister toolkit for model explainability. Many production teams pair the two so that a flagged fairness violation can be investigated with explanation methods like Contrastive Explanations or LIME.
- **IBM Watson Studio / watsonx**: AIF360 metrics are exposed inside IBM's Watson data-science platform as part of the Trusted AI lifecycle.
- **[MLflow](https://aiwiki.ai/wiki/mlflow)**: community examples log AIF360 metrics as MLflow custom metrics so they are tracked alongside accuracy and AUC.

## How does AIF360 compare with other fairness toolkits?

A handful of open-source fairness toolkits emerged in the same 2018-2020 window, each with slightly different priorities. The table below summarises the main options.

| Toolkit | Organisation | Language | Licence | Primary focus | Mitigation algorithms | Notable feature |
|---------|--------------|----------|---------|---------------|------------------------|------------------|
| AIF360 | IBM, now LF AI & Data | Python and R | Apache 2.0 | End-to-end metrics and mitigation | 15 listed in the README, across pre, in, post | Largest library of mitigation algorithms |
| Fairlearn | Started at [Microsoft](https://aiwiki.ai/wiki/microsoft), now community-driven | Python | MIT | Reductions approach plus disaggregated metrics | One preprocessing method, reductions and adversarial in-processing, threshold-optimizer post-processing | MetricFrame plotting (its Jupyter dashboard widget was removed in version 0.7.0) [33] |
| Aequitas | University of Chicago | Python and CLI | MIT | Audit reports for policymakers | Mitigation methods added via Aequitas Flow in version 1.0.0 (2024) | Bias audit reports and fair-ML experimentation [34] |
| TF Fairness Indicators | [Google](https://aiwiki.ai/wiki/google) | Python (TFX) | Apache 2.0 | Slice-based metrics for [TensorFlow](https://aiwiki.ai/wiki/tensorflow) models | None | Tight integration with TFX pipelines |
| What-If Tool | Google PAIR | Python ([TensorBoard](https://aiwiki.ai/wiki/tensorboard)) | Apache 2.0 | Interactive what-if analysis | None; counterfactual exploration | Visual interactive tool |
| Themis-ML | N. Bantilan 2017 | Python | MIT | Early academic toolkit | Relabelling, an additive counterfactually fair estimator, and reject option classification | Predates AIF360 |

In practice many teams use more than one. A common workflow is to audit with Aequitas because it produces a report that lawyers and policy people can read, mitigate with AIF360 because it has the widest selection of methods, and visualise with the What-If Tool when a senior stakeholder wants to interrogate individual records.

## What are the limitations of AIF360?

The academic literature has been clear that no toolkit can resolve the fundamental tensions in the field.

**The impossibility theorems.** Chouldechova (2017) and Kleinberg, Mullainathan, & Raghavan (2017) independently showed that several intuitive fairness criteria, including calibration, predictive parity, and equalised false-positive rates, cannot all hold simultaneously when base rates differ across groups [22][23]. AIF360 implements the metrics but cannot square this circle: a user has to pick which fairness criterion they care about, and that choice is normative, not technical.

**Group fairness versus individual fairness.** Most AIF360 metrics are group fairness metrics. They say nothing about whether two similar individuals from different groups are treated similarly, which is the [individual fairness](https://aiwiki.ai/wiki/individual_fairness) criterion of Dwork et al. (2012) [24]. Consistency and Theil index partially address this, but the toolkit's centre of gravity is group fairness.

**Binary protected attributes.** The bulk of the implementations assume a single binary protected attribute. Intersectional fairness (Crenshaw 1989; Buolamwini & Gebru 2018) is supported through the differential fairness metrics added later, but in-processing algorithms generally do not natively handle multiple intersecting protected attributes [28]. The GerryFair classifier is the main exception.

**Need for labelled protected attributes.** Most mitigation algorithms require the protected attribute to be available at training time. In many production settings this is legally or ethically constrained (the GDPR places restrictions on processing of "special category" personal data), and "fairness through unawareness" is known to be insufficient. AIF360 cannot solve this; it assumes the data is already there.

**Performance loss is not always quantified.** Most published fairness mitigations reduce model accuracy or revenue. The toolkit reports both the fairness metric and standard accuracy after mitigation, but it does not always make the Pareto trade-off explicit, and the optimal point on that trade-off is application-specific.

**Cultural and regulatory context.** The four-fifths rule is from the US EEOC and has no force in the EU, the UK, India, or most other jurisdictions [25]. AIF360 ships it as a metric without enforcing this caveat in code. Users who deploy AIF360 outside the US should not assume that passing the four-fifths rule means they are compliant with local anti-discrimination law.

**It does not replace bias auditing by humans.** Friedler et al. (2019) and Holstein et al. (2019) found that the social and organisational context in which a model is deployed often dominates the technical fairness intervention [26][27]. AIF360 is a useful tool inside that process, not a substitute for it.

## How widely is AIF360 used?

AIF360 has been cited by thousands of academic papers since 2018 and is widely used as a teaching tool in machine-learning ethics courses. The toolkit is the reference implementation for several follow-up fairness papers, which makes raw citation counts somewhat misleading: a paper that benchmarks against AIF360 cites it even if the paper proposes a competing method.

Industry usage is harder to measure because compliance pipelines are rarely public, but several documented examples exist. IBM uses AIF360 inside its watsonx Governance product as part of the Trusted AI lifecycle. The dataset and mitigation choices in some published audit reports, including for the COMPAS dataset, mirror the AIF360 pipeline.

The project is also part of IBM's broader Trusted AI suite, which together with AIF360 includes:

- **AIX360** (AI Explainability 360) for model explanations.
- **ART** (Adversarial Robustness Toolbox) for adversarial example defences.
- **Uncertainty Quantification 360** for predictive uncertainty estimation.
- **AI FactSheets 360** for model documentation.

## Recent developments

The most recent tagged releases are version 0.6.0 (February 2024), which added the SenSeI/SenSR individual-fairness algorithms, deterministic re-ranking for fair ranking in search and [recommender system](https://aiwiki.ai/wiki/recommender_system)s, and an optimal-transport distance metric, and version 0.6.1 (April 2024), which added the FACTS subgroup counterfactual detector [35]. The MDSS bias subset scanning method, based on the multi-dimensional subset scan of Zhang and Neill (2017), had been added in version 0.5.0 (September 2022) to support intersectional bias detection [3][35]. As of November 2026 no newer release has been tagged, but the main branch has continued to receive changes, including an intersectional bias mitigation contribution in October 2025 and security fixes for model loading in June 2026, and the README lists support for Python 3.10 to 3.13 on macOS, Ubuntu, and Windows [3].

Community discussion under LF AI & Data takes place on an open Slack workspace [3], and each release's notes credit new contributors from outside the original IBM team [35].

## ELI5: AI Fairness 360 in plain terms

Imagine a computer program that decides who gets a loan. If the program quietly says "yes" more often to one group of people than another, that is unfair bias. AI Fairness 360 is a free toolbox that does two things. First, it measures how lopsided the program's decisions are, using more than 70 different fairness rulers. Second, it offers more than a dozen different ways to fix the lopsidedness: it can clean up the data before the program learns from it, change how the program learns, or adjust the program's answers after the fact. IBM built it and then gave it away to a non-profit so anyone can use it for free.

## See also

- [Algorithmic fairness](https://aiwiki.ai/wiki/algorithmic_fairness)
- [Fairness metric](https://aiwiki.ai/wiki/fairness_metric)
- [Disparate impact](https://aiwiki.ai/wiki/disparate_impact)
- [Individual fairness](https://aiwiki.ai/wiki/individual_fairness)
- [Incompatibility of fairness metrics](https://aiwiki.ai/wiki/incompatibility_of_fairness_metrics)
- [AI bias](https://aiwiki.ai/wiki/ai_bias)
- [Counterfactual fairness](https://aiwiki.ai/wiki/counterfactual_fairness)
- [IBM](https://aiwiki.ai/wiki/ibm)

## References

1. Bellamy, R. K. E., Dey, K., Hind, M., Hoffman, S. C., Houde, S., Kannan, K., Lohia, P., Martino, J., Mehta, S., Mojsilovic, A., Nagar, S., Ramamurthy, K. N., Richards, J., Saha, D., Sattigeri, P., Singh, M., Varshney, K. R., & Zhang, Y. (2018). "AI Fairness 360: An Extensible Toolkit for Detecting, Understanding, and Mitigating Unwanted Algorithmic Bias." arXiv:1810.01943. https://arxiv.org/abs/1810.01943
2. Bellamy, R. K. E. et al. (2019). "AI Fairness 360: An Extensible Toolkit for Detecting and Mitigating Algorithmic Bias." *IBM Journal of Research and Development*, vol. 63, no. 4/5. https://ieeexplore.ieee.org/document/8843908
3. Trusted-AI. "AIF360 GitHub Repository." https://github.com/Trusted-AI/AIF360
4. AIF360 Documentation. https://aif360.readthedocs.io/
5. IBM Research Blog. "Introducing AI Fairness 360." 19 September 2018. https://research.ibm.com/blog/ai-fairness-360
6. LF AI & Data Foundation. "3 Trusted AI Toolkits Join LF AI as Newest Incubation Projects." 22 September 2020. https://lfaidata.foundation/blog/2020/09/22/3-trusted-ai-toolkits-join-lf-ai-as-newest-incubation-projects/
7. Calders, T., & Verwer, S. (2010). "Three naive Bayes approaches for discrimination-free classification." *Data Mining and Knowledge Discovery*, 21(2), 277-292.
8. Hardt, M., Price, E., & Srebro, N. (2016). "Equality of Opportunity in Supervised Learning." *NeurIPS 2016*. arXiv:1610.02413.
9. Pleiss, G., Raghavan, M., Wu, F., Kleinberg, J., & Weinberger, K. Q. (2017). "On Fairness and Calibration." *NeurIPS 2017*.
10. Kamishima, T., Akaho, S., Asoh, H., & Sakuma, J. (2012). "Fairness-Aware Classifier with Prejudice Remover Regularizer." *ECML PKDD 2012*.
11. Zhang, B. H., Lemoine, B., & Mitchell, M. (2018). "Mitigating Unwanted Biases with Adversarial Learning." *AAAI/ACM AIES 2018*.
12. Kamiran, F., & Calders, T. (2012). "Data Preprocessing Techniques for Classification without Discrimination." *Knowledge and Information Systems*, 33(1), 1-33.
13. Calmon, F., Wei, D., Vinzamuri, B., Ramamurthy, K. N., & Varshney, K. R. (2017). "Optimized Pre-Processing for Discrimination Prevention." *NeurIPS 2017*.
14. Feldman, M., Friedler, S. A., Moeller, J., Scheidegger, C., & Venkatasubramanian, S. (2015). "Certifying and Removing Disparate Impact." *KDD 2015*.
15. Zemel, R., Wu, Y., Swersky, K., Pitassi, T., & Dwork, C. (2013). "Learning Fair Representations." *ICML 2013*.
16. Celis, L. E., Huang, L., Keswani, V., & Vishnoi, N. K. (2019). "Classification with Fairness Constraints: A Meta-Algorithm with Provable Guarantees." *FAT* 2019*.
17. Agarwal, A., Beygelzimer, A., Dudik, M., Langford, J., & Wallach, H. (2018). "A Reductions Approach to Fair Classification." *ICML 2018*.
18. Kamiran, F., Karim, A., & Zhang, X. (2012). "Decision Theory for Discrimination-Aware Classification." *ICDM 2012*.
19. Kearns, M., Neel, S., Roth, A., & Wu, Z. S. (2018). "Preventing Fairness Gerrymandering: Auditing and Learning for Subgroup Fairness." *ICML 2018*.
20. Speicher, T., Heidari, H., Grgic-Hlaca, N., Gummadi, K. P., Singla, A., Weller, A., & Zafar, M. B. (2018). "A Unified Approach to Quantifying Algorithmic Unfairness." *KDD 2018*.
21. Foulds, J. R., Islam, R., Keya, K. N., & Pan, S. (2020). "An Intersectional Definition of Fairness." *ICDE 2020*.
22. Chouldechova, A. (2017). "Fair Prediction with Disparate Impact: A Study of Bias in Recidivism Prediction Instruments." *Big Data*, 5(2), 153-163.
23. Kleinberg, J., Mullainathan, S., & Raghavan, M. (2017). "Inherent Trade-Offs in the Fair Determination of Risk Scores." *ITCS 2017*.
24. Dwork, C., Hardt, M., Pitassi, T., Reingold, O., & Zemel, R. (2012). "Fairness Through Awareness." *ITCS 2012*.
25. US Equal Employment Opportunity Commission. (1978). "Uniform Guidelines on Employee Selection Procedures." 29 CFR Part 1607.
26. Friedler, S. A., Scheidegger, C., Venkatasubramanian, S., Choudhary, S., Hamilton, E. P., & Roth, D. (2019). "A Comparative Study of Fairness-Enhancing Interventions in Machine Learning." *FAT* 2019*.
27. Holstein, K., Wortman Vaughan, J., Daume III, H., Dudik, M., & Wallach, H. (2019). "Improving Fairness in Machine Learning Systems: What Do Industry Practitioners Need?" *CHI 2019*.
28. Buolamwini, J., & Gebru, T. (2018). "Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification." *FAT* 2018*.
29. AI Business. "IBM donates AI fairness and explainability tools to the Linux Foundation." 2020. https://aibusiness.com/companies/ibm-donates-ai-fairness-and-explainability-tools-to-the-linux-foundation
30. Angwin, J., Larson, J., Mattu, S., & Kirchner, L. (2016). "Machine Bias." *ProPublica*, 23 May 2016. https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing
31. Deutscher, M. (2020). "IBM donates machine learning tooling to enable 'responsible' AI." *SiliconANGLE*, 29 June 2020. https://siliconangle.com/2020/06/29/ibm-donates-machine-learning-tooling-enable-responsible-ai
32. CRAN. "Package aif360" (archived 20 June 2023). https://cran.r-project.org/web/packages/aif360/index.html
33. Fairlearn. "v0.7.0" release notes. https://fairlearn.org/main/user_guide/installation_and_version_guide/v0.7.0.html
34. Data Science for Social Good. "Aequitas: Bias Auditing & 'Correction' Toolkit" (GitHub README). https://github.com/dssg/aequitas
35. Trusted-AI. "AIF360 Releases." GitHub. https://github.com/Trusted-AI/AIF360/releases

