May 29, 2024

Impact of EU AI Act on Skills-based Intelligence

Impact of EU AI Act on Skills-based Intelligence

Here at SkyHive by Cornerstone, we welcome the EU AI Act as validation of SkyHive’s commitment to responsible and ethical AI. 

In this article, we discuss the EU AI Act and its relevance for AI-powered skills-based intelligence. 

The EU AI Act may also influence regulation in other regions, including the AI Safety Institute Consortium in the United States and the Guide on AI Governance and Ethics by the Association of Southeast Asian Nations (ASEAN). 

About the EU AI Act

The EU AI Act definition of AI mirrors the OECD AI definition. AI systems operate with some level of autonomy and infer from inputs how to generate content, predictions, and recommendations that influence virtual or physical environments. 

The AI Act builds upon the EU’s previous GDPR mandates for data privacy. 

The Act applies to organizations and individuals in the European Union who deploy or import AI systems. Excluded are AI systems developed or used exclusively for military purposes. 

Despite a length of several hundred pages, there are some aspects of the Act that remain unclear. “There is little clarity and precision regarding [how the EU AI Act covers] distributors”, notes law firm WilmerHale.

The EU AI Act provides for EU-wide rules on data quality, transparency, human oversight, and accountability. Fines for noncompliance range up to 35 million Euros or 7% of global annual revenue, whichever is higher. 

The European Parliament adopted the AI Act on March 13, 2024. The Act enters into force 20 days after publication in the Official Journal. What comes next is a phased implementation between late 2024 and mid 2026. Bans on prohibited AI systems start in six months.  Obligations for general-purpose AI such as GenAI large language models (LLMs) start in mid 2025. 

KPMG provides a helpful chronology in “Decoding the EU AI Act”. 

Starting in April 2021 when the European Commission unveiled a proposal for a new AI Act, the EU’s legislative journey has gone through multiple stages to reach the final text in the first half of 2024 and complete enforcement starting Spring 2026.

The EU AI Act establishes a four-part range of risk: unacceptable, high, limited, and minimal risks. 

  • Unacceptable risks include social scoring by public authorities. Unlike skills-based intelligence, which helps match people with relevant skills to available jobs, this prohibition applies to profiling or discrimination by social behavior. 
  • High risk AI systems are permitted but must meet the most stringent requirements. The EU AI Act includes analysis of job applications and evaluation of job candidates as high risk. 
  • Limited risk covers use of general purpose AI such as GenAI models that facilitate natural language questions of skills-intelligence data.
  • Minimal risk includes email spam filters and video games. “What qualifies as low-risk remains unclear” per Mercer.

The EU AI Act describes four levels of risk: minimal, limited, high, and unacceptable, with examples of what falls under each level.

Source for image: KPMG “Decoding the EU AI Act

SkyHive Compliance with the EU AI Act

The Act’s provisions require that AI systems minimize bias that can result in unfair or inadequate outcomes. This is near and dear to SkyHive. 

For SkyHive, we view compliance with the EU AI Act and GDPR as integral to our longstanding commitment to responsible and ethical AI. 

SkyHive Skill Models are Armilla Verified as free of AI bias and meet the demanding standards set by the EU AI Act, New York City’s Local Law 144, and future regulations.

SkyHive ethical AI has been vetted or audited by over 100 large enterprise and government customers. SkyHive Co-Founder and CTO Mohan Reddy serves as an expert advisor for the Responsible AI Institute.

Our data collection, processing, and privacy protections follow six principles to ethical and responsible AI: transparency, explainability, robustness, trust, confidentiality, and accountability. 

Powering the Transition from Jobs to Skills 

As the leader in AI-powered skills intelligence, SkyHive supports organizations and communities to hire, manage & retain people with in-demand skills and to upskill and reskill your workforce. 

SkyHive is transitioning the European fishing and aquaculture sector with financial backing from the European Institute of Innovation and Technology, a body of the European Union. 

SkyHive profiles go beyond resumes, CVs, and LinkedIn profiles to include education, hobbies, and credentials. 

When people self-report on their skills, they impose limitations. They don’t fully realize how skilled they are. On average, individuals identify 11 skills for their particular role. Using SkyHive, that number jumps to an average of 34. 

Identifying a wider range of skills, and adjacent skills available to learn, has important benefits for professional development, performance management, learning and development.

In awarding SkyHive one of the “Next Big Technologies Working for Social Good in 2023”, Fast Company explains that SkyHive “uses AI to match people to jobs they might not have thought were a fit but that they actually have the transferable skills for…. and it even shows learning opportunities that could help bridge a skill gap to enter a new career.”

SkyHive Explainable AI 

Mercer advises to “Avoid building and using high-risk tech solutions, such as ‘black-box’ AI tools that automate HR processes with little documentation and transparency. There is a risk that these may be banned or difficult to implement, given the EU AI Act.” 

In contrast to “black-box” AI tools, SkyHive adopts explainable AI and human-in-the-loop approaches, so that humans can understand and interpret the model outputs. 

SkyHive Labor Market Intelligence (LMI) provides global data on job vacancies and in-demand skills in real time. SkyHive ingests an average of 28TB or more of raw data a day across 200 countries and territories.  

The SkyHive approach combines skill extraction and inference with LMI-based recommendations. We produce the market’s largest knowledge graph of human capital data. This includes over 5 billion job descriptions and 1 billion anonymized job profiles. 

SkyHive never uses customer data to train our models. We gather public-domain data on the labor market and in-demand skills from job boards, resumes, course outlines, census data, corporate annual reports, and government labor market statistics. 

The LMI looks at patent applications and academic journals to derive new skills that are becoming in demand. 

For more details on SkyHive data sources, visit the website FAQ page

About SkyHiveGPT 

At SkyHive, we have developed our own Large Language Model (LLM) called SkyHiveGPT. Our AI system is built upon algorithms and models that we have developed in-house. SkyHiveGPT is a domain fine-tuned and pre-trained LLM. The models are continuously monitored for improvement and mitigation of AI basis. 

Our dedicated team continuously monitors the model's outputs, collects feedback from human reviewers and end-users, and works on improving the model's performance. 

This involves fine-tuning the model on additional data, adjusting hyperparameters, retraining the model or refining the prompt engineering strategies. 

SkyHive uses Transformer-based deep learning models as the core of the technology stack, leveraging their powerful natural language processing capabilities to analyze and understand large volumes of workforce-related data. 

SkyHive applies these models to tasks such as parsing job descriptions, extracting relevant skills, and matching job requirements with employee skill profiles.

SkyHive Mitigation of AI Bias

SkyHive employs different bias mitigation methods depending on the situation, including for machine translation, sentiment analysis, language models, and word embedding analogies.

We employ:

  • Algorithms for discrimination discovery.
  • Discrimination prevention by means of fairness-aware data.
  • Human in the loop approach.

The framework for fairness-aware modeling includes the following:

  • Baseline: Train a model on all available input variables in the source dataset, including protected attributes. 
  • Remove Protected Attribute: Train a model on input variables without protected attributes. This is what’s referred to as a naive fairness-aware approach. 
  • Relabel Target Variable: Train a model using the Relabeling fairness-aware method. 
  • Counterfactually Fair Model: Train a model using the Additive Counterfactually Fair method. 
  • Reject-option Classification: Train a model using the Reject-option Classification method.

SkyHive Commitment to Our European Customers 

SkyHive announced an expansion of European operations in November 2023. Over 30 percent of SkyHive’s customer base is headquartered in Europe. We are staffing several new roles in Dublin

About SkyHive by Cornerstone

Recognized by the World Economic Forum and Gartner for our contributions to the future of work, SkyHive enables organizations and communities worldwide from Best Buy, the Government of Canada, and Collège La Cité to Unilever and Zinnia transition from jobs to skills, backed by partnerships including Accenture, SAP, and Workday and non-profits Opportunity@Work and JobsFirstNYC. 

Our products Skill Passport™, SkyHive Enterprise, SkyHive Platform, and the foundational Human Capital Operating System™ support individuals to upskill and reskill while enabling organizations and communities to hire, manage, and retain people with in-demand skills. Learn more about how to unleash human potential at

Learn more about how to unleash human potential at

Cornerstone OnDemand Inc. acquired SkyHive on May 22, 2024. See the press release for the exciting news. Cornerstone Galaxy, the complete AI-powered Workforce Agility platform, allows organizations to identify skills gaps and development opportunities, retain and engage top talent, and 
provide multi-modal learning experiences to meet the diverse needs of the modern workforce. Learn more at

Download PDF

Ready to unleash potential across your workforce?

The world's most ethical Al people technology to help you transition from jobs-based to skills-based — award-winning, demonstrated, and internationally recognized.

Request a demo

Related resources

By clicking “Accept All Cookies”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts. View our Privacy Policy for more information.