The Ethical AI Database (EAIDB) is a curated collection of startups that are either actively trying to solve problems that AI and data have created or are building methods to unite AI and society in a safe and responsible manner.
2022 was a big year for our recognized startups. We witnessed multiple strategic acquisitions, lots of dynamic movement within each category, and a ton of investment inflow. EAIDB was founded in Q2 2022 with 148 companies, but since then we have grown by about 45% and now have 215 verified companies. We continue to capture more of the space every day. View our growing market map here.
This report is a conglomeration of insights we have gained throughout the course of 2022. We cover market movements within each of our five categories, highlights for the year, and more. EAIDB publishes reports and market maps on a semiannual basis (switching away from quarterly reports to provide more holistic and interesting analyses).
We welcome 25 new firms to the ethical AI ecosystem. The database now contains 216 companies actively working to better the way we use artificial intelligence and machine learning in practice. Welcome to EAIDB!
Note: if you have an AdBlocker running, the logos below may not display.
We have also identified some companies in EAIDB that we feel no longer fit our criteria upon closer inspection. Some are simply too matured (we have a strong preference for startups that are fairly early on in their journey, Series E or earlier), some are inactive, and some we feel do not directly relate to AI ethics or its many subcategories. As we continue to refine our criteria, EAIDB will always be transparent in the changes made to its map and constituents. The following companies were removed from the database:
The distribution of companies in EAIDB is skewed heavily towards vertical solutions, primarily because of the more mature hiretech and fintech markets. The other categories are just as relevant, however, and tackle different aspects of the AI pipeline.
Ethical AI startups raised over $1.1bn in 2022 from 55 companies. The most common round was a Series A investment.
|Company||Amount (millions USD)||Round||Lead Investors|
|Mission Control||2.0||Pre-Seed||Stage Venture Partners|
|Veil AI||1.3||Pre-Seed||BioInnovation Institute|
|Anch AI||2.1||Seed||Benhamou Global Ventures|
|Black AI||5.4||Seed||Jelix Ventures|
|Checkstep||5.0||Seed||Dawn Capital, Form Ventures|
|HiddenLayer||6.0||Seed||Ten Eleven Ventures|
|Pendella||5.2||Seed||American Family Ventures, MassMutual Ventures|
|Privya AI||6.0||Seed||Hyperwise Ventures|
|Protopia AI||6.0||Seed||ATX Venture Partners|
|Alva Labs||13||Series A||VNV Global|
|Aporia||25||Series A||Tiger Global Management|
|Bodyguard AI||11||Series A||Keen Venture Partners, Ring Capital|
|Brighter AI||Undisclosed||Series A||Armilar Venture Partners|
|CausaLens||53||Series A||Molten Ventures, Dorilton Capital|
|Credo AI||13||Series A||Sands Capital Ventures|
|Diversio||6.3||Series A||First Round Capital|
|FairPlay AI||10||Series A||Nyca Partners|
|Mathison||25||Series A||F-Prime Capital|
|Modulate AI||30||Series A||Lakestar|
|Private AI||8.0||Series A||BDC Venture Capital|
|Sapia||11.8||Series A||Macquarie Partners, W23|
|Synthesis AI||17||Series A||468 Capital|
|X0PA AI||4.2||Series A||ICCP Venture Partners|
|Arize AI||38||Series B||TCV|
|Arthur AI||42||Series B||Acrew Capital, Greycroft|
|Cohere||122||Series B||Tiger Global Management|
|Datagen||50||Series B||Scale Venture Partners|
|Mostly AI||25||Series B||Molten Ventures|
|Spectrum Labs||32||Series B||Intel Capital|
|TruEra||25||Series B||Menlo Ventures|
|Flock Safety||150||Series E||Tiger Global Management|
|MDClone||63||Series C||Viola Growth, Warburg Pincus|
|Pave||100||Series C||Index Ventures|
|Ambient AI||52||Undisclosed||Andreessen Horowitz|
|Enzai Technologies||0.7||Undisclosed||Techstart Ventures|
|Pipeline Equity||Undisclosed||Undisclosed||Workday Ventures|
|QuadFi||99||Debt Financing||Crayhill Capital Management|
|Troj AI||2.3||Undisclosed||Build Ventures, Flying Fish Partners|
There were three separate acquisitions made by companies in EAIDB. All three were made by hiretech companies (Crosschq, Pave, Alva Labs) on the back of their most recent funding rounds.
Crosschq (candidate insight and people analytics platform with DE&I tracking) acquired TalentWall (data-driven recruiting analytics platform) following their $30m Series A funding round.Read PR
Pave (compensation benchmarking) acquired Advanced-HR (similar platform) following their $100m Series C funding round. They have integrated Advanced-HR's "Option Impact" product into their own offering.Read PR
Alva Labs (candidate assessment platform) acquired DevSkills (online coding skills testing platform) in order to provide their highly customizable coding tests through their platform.Read PR
There were also four exits within EAIDB's ecosystem. Again, most of these were hiretech related (since fair hiretech is the most mature of all verticals in the ethical AI ecosystem). Since the "hiretech boom" of the mid-2010s, this area has been gaining momentum and is finally beginning to show investors some dividends.
Paycor acquires Talenya to integrate their AI-based recruiting software to their existing human capital management (HCM) platform. Talenya, an EAIDB startup, is a powerful tool for tracking and executing companies' DE&I strategies.Read PR
In another hiretech acquisition, Silverback United acquires Headstart AI, an automated hiring platform to perpetuate their strategy of building a portfolio of vertically-specific, high data-value companies.Read PR
Reddit acquihires Oterlu AI to bolster their content moderation team. Oterlu provides automated content moderation technology, but Reddit was after the talent within the Oterlu team to apply their thought leadership to Reddit's existing technology.Read PR
The only non-hiretech acquisition within EAIDB's universe in 2022, Anonos (data privacy / security) acquired Statice (synthetic data platform). Statice's technology will be integrated into Anonos' "Data Embassy" product.Read PR
There was a ton of innovation in 2022 from the ethical AI ecosystem. From the launch of new companies to the introduction of groundbreaking technologies, the ecosystem has never looked so impressive. Here are a few highlights from the year.
Neurosymbolic AI is the future. Umnai (founded in 2019), a leading researcher and technology provider of neurosymbolic AI, has developed technology that significantly improves existing ML methodologies with inherently auditable, interpretable, and faster models. The company gave several presentations in 2022, including at AI Forum. Read more about their technology here.
Integrate AI has developed a seamless way to perform federated learning. Federated learning provides strong privacy guarantees because it is a methodology that never allows data to leave its silo. Integrate has created developer-friendly APIs to access their technology and launched their federated learning platform in 2022.
Superwise operationalizes model observability. Their Model Observability platform was made available on the Datadog marketplace in March 2022. This is essentially a marker of product differentiation and tried-and-true product quality.
Ethically Aligned AI is a consulting / educational firm. In 2022, they developed a microcredential named "Artificial Intelligence Ethics" with Canada's Athabasca University. More information can be found here.
FairPlay AI debiases lending algorithms to produce better, more fair outcomes. Their "Mortgage Fairness Report" on the state of mortgage fairness in the United States was released in late 2022. More information can be found here.
Cangrade focuses on bias reduction in hiring decisions. In 2022, the company was granted a patent on their particular method of bias removal via a separate "Adverse Impact" test that is applicable outside of just hiring. Read more here.
QuadFi is a lending platform with support for thin or no files. In 2022, the company released a product for global immigrants with little to no credit history (and does not rely on FICO scores). Read more here.
Aporia is a model monitoring and observability company. Last year, they launched their novel Direct Data Connectors (DDC) service, which allows one to directly connect to training and inference datasets to essentially be able to monitor predictions without sampling, production code changes, or cloud costs. Read more here.
Modulos AG is a data debiasing service. In 2022, they released their data-centric platform that iteratively debiases data with a human-in-the-loop postprocess. Read more here.
Monitaur operationalizes AI governance. The company released a new product, "GovernML," allows one to maintain a system of record of model governance policies, ethical practices, and model risks. Find the press release here.
Each category in EAIDB represents a different "type" of ethical AI service and therefore display very different dynamics over the course of a year. Below are some highlighted trends that seem to be driving the industry forward.
For more information on each category, visit our methodology.
It takes a village to support an open-source framework and keep it updated. Existing packages on the internet today lag behind the most current technology and are often not enough on their own to build truly responsible applications.
The following chart represents the average number of commits for several well-known open-source packages on Github. In blue are general AI/ML libraries; in red are responsible development toolkits. Notice the stark contrast in human capital.
But this is not something unexpected. These packages are necessary and fulfil a crucial role but cannot keep up with the human capital delivered to AI-enabling developer tools. These libraries are undermaintained because there are too few resources allocated to the problem of ethics in AI/ML. This may be changing, however, as some open-source technologies like BigScience Bloom and ChatGPT alternative PaLM + RLHF have reignited buzz around open-source developer tools. In the meantime, attention and resources continue to be delivered to AI enablers, not AI problem-solvers.
With companies like Arthur AI and Fiddler AI paving the way, this category of startups seems to be growing the fastest relative to the others. MLOps companies are reaching Series A investment rounds much faster on average, despite being earlier in its lifecycle relative to other categories like hiretech or data for AI companies.
The chart on the left depicts the proportion of each category / subcategory in EAIDB that have reached at least a Series A funding round. The chart on the right shows the general lifecycle of companies founded from 2015-2021 in each category. Note that the MLOps curve reached its peak much later than the others; this means the MLOps category has newer companies that have already exceeded growth expectations from a funding perspective.
This could be an indication that the investment world still views responsible AI as primarily a developer-centric problem. AI GRC companies, which typically come into play after models are built, seem to be struggling to prove (to investors specifically) that they are needed.
In 2022, one of the earliest acquisitions within EAIDB's "Data for AI" category was made. Anonos, a data protection company specializing in PII obfuscation and privacy preservation, acquired Statice GmbH, a German synthetic data company.
|Company||Focus||Status||Last Funding Round|
|Statice GmbH||finance, healthcare, insurance||Acquired||Seed|
|Mostly AI||finance, insurance||Active||Series B|
Statice was unique in that they were really more of a targeted synthetic data company. In essence, they chose to focus their efforts in the financial vertical (and won some awards for it as well) and built functionality to cover every use case. They were specialists in a generalist space. Statice is one of the only early-stage synthetic data companies that do this. From Anonos' perspective, this makes Statice far more appealing than its competitors because Anonos' clients primarily operate in "financial services, media, and pharmaceuticals". In Anonos' words, "Statice's vision for data agility combined with privacy reliability changed how we drive digital transformation for our clients." Perhaps this is an indication that, in the responsible development sphere, it is better to dive vertically rather than spread horizontally if entering the market late.
In cases where data privacy is of the utmost importance, methods of machine learning that access the data and bring it into a centralized platform (a warehouse, database, etc.) are typically considered risky, even with cutting-edge security. Federated learning, productionized by companies like Integrate AI allow algorithms to learn local methods over each dataset and report parameters back to a global model. Because of this scheme, the global model (the one put into production) has never seen a single row of the original data.
Spearheaded by firms like CausaLens, Causal AI is a branch of machine learning that takes causal effect as the primary learning mechanism instead of correlations. This makes causal AI inherently interpretable and can generate insights on a much more robust level. There are, of course, caveats with this approach, but the idea that all decisions are immediately auditable makes causal AI an attractive prospect.
The newest of these three technologies (and arguably the most difficult to understand!), neurosymbolic AI is a combination of neural and logic-based symbolic architectures. The output of these models are neural-type networks that are inherently explainable, auditable, and interpretable. Companies like UMNAI are currently leading the charge.
These technologies, however, are still in their early stages. Though the AI climate is relatively warm to causal AI, it has not quite seen the full potential of federated learning or neurosymbolic AI. These paradigm shifts usually take years to really accelerate as switching costs away from traditional methods are excessively high for pre-existing applications. Gartner's Hype Cycle cites causal AI as likely to reach peak adoption in the next two to five years.
This category of startups includes any vertically-oriented startup (hiretech, fintech, etc.) as well as companies forming horizontally-oriented technologies (e.g. a company like AlgoFace which develops facial detection technology for any use case).
Consulting firms are generally considered lighthouses in the murky world that is AI ethics. With so many new frameworks, technologies, etc., many clients are understandably overwhelmed. They often turn to consulting firms to provide guidance. Some of the larger firms in the consulting space (the Deloittes, Accentures, McKinseys of the world) are often used as a first resort but often don't have the specific expertise necessary to solve the problem. The gap they have left in the market left a massive pain point for customers, which created an inflow of new firms over the last few years.
Note that EAIDB's discovery process lags new firm creation. The trend from 2018-2020 is expected to hold true in 2022.
Based on the patterns of innovation and momentum observed throughout 2022, EAIDB presents the following high-level insights and predictions.
With such a large market gap in the open-source space, EAIDB expects major contributions from organizations well-poised to deliver impactful software directly to individuals. These organizations might be universities / research centers or governmental bodies. New open-source products, frameworks, and initiatives are expected to hit the market by 2H2023.
Generalists (i.e., context-agnostic products) in this market typically lose to specialists (not a fixed rule, but holds empirically true). The name of the game in the ethical AI ecosystem is specificity, especially for newer companies. The depth of a product or service and its use case coverage within a vertical will play a critical role in how the market decides winners and losers. As a means of product differentiation, many companies will adapt their technology to offer better coverage for their first few clients and will therefore transform into a specialist firm. This is one theory as to why Statice GmbH was so coveted by Anonos (in addition to their outstanding proprietary technology).
Hiretech (which boomed in the mid-2010s) is already slowing in terms of new companies founded because the subcategory has reached a stage of maturity. We are currently witnessing a lot of M&A activity in the particular subsector of "ethical hiretech." This may continue as the market consolidates a bit. Hiretech will soon be outpaced by other emerging sectors like text/vision, insurtech, healthtech, etc. Data privacy is expected to be the "next hiretech" as it only slightly lags hiretech in maturity. We have already seen M&A activity within the data privacy sector, and this is only expected to grow in momentum.
In 2022, EAIDB partnered with two institutions to perpetuate knowledge and awareness of the ethical AI ecosystem and why it is so critical in today's automated world.
EAIDB and Nordic Innovation collaborated on a "Nordic Ethical AI Map" (viewable here) to increase awareness of organizations that are actively working to improve the way AI is built. Nordic Innovation is responsible for fostering cross-border trade and innovation in the Nordic region.
EAIDB and the Montreal AI Ethics Institute (MAIEI) collaborated on an ethical AI series covering each of the categories in the database. MAIEI regularly publishes content related to AI ethics.
The Ethical AI Database is a live database of curated startups attempting to solve some of the most damaging aspects of AI / ML in society. We offer semiannual market map updates and reports, but periodically release content through various media channels. For more on how we curate our database, view our methodology. To submit your company to our list, fill out the submission form. If you'd like to work with us, you can reach out here. Thank you to the Ethical AI Governance Group (EAIGG) for assisting us with our mission.
Disclaimer: logos were taken from LinkedIn company profiles or were found via search engines, but belong to their respective firms.