Artificial Intelligence: Judgment and Standards Make the Difference

Why the Testing and Certification of AI Systems Can No Longer Be Postponed

-Article by SGS Romania (ARILOG member)-

The Paradox of the Artificial Intelligence Era

We live in an era obsessed with artificial intelligence. We have built machines capable of processing, in mere seconds, volumes of information that entire generations could not get through in a lifetime. From medical diagnosis to financial strategies, AI has become the new oracle—an omnipresent cognitive partner whispering answers with dizzying speed and self-assurance.

And yet, a paradox is becoming increasingly evident: despite speeds inconceivable just a few years ago, the quality of judgment appears to be at significant risk. This disconnect reveals a fundamental confusion we have cultivated with the arrogance characteristic of modernity: we have conflated intelligence with discernment. We erroneously assumed that superior data-processing capability equates to superior reasoning ability. Today, at the dawn of the new AI era, this confusion has become our greatest strategic risk.

Intelligence versus Discernment

Intelligence, in its pure form, is the capacity to process information: speed, pattern recognition, and the construction of logical connections. A Large Language Model (LLM) represents the pinnacle of this form of intelligence: it can summarize, translate, code, and generate text that, on the surface, appears coherent and informed.

Discernment, on the other hand, is the capacity to judge the meaning, relevance, and consequences of information. It is not about what you know, but about how you apply what you know within a context that is real, complex, and—more often than not—ambiguous. Discernment is the art of consciously deciding what to ignore, when to stop, and when to question even the most elegant conclusions. It is a faculty born not of data, but of experience, of an understanding of human nature, and of an awareness of the stakes and responsibilities involved. Aristotle called this faculty *phronesis*—practical wisdom. Kant spoke of practical reason, which requires the engagement of the will in the world. Discernment unites cognition with experience—a faculty that machines, by their very nature, cannot authentically claim.

When AI Gets It Wrong – Costly Lessons

The specific cases below offer lessons regarding the dangers of unchecked delegation of tasks to systems that lack consciousness.

In 2024, a Canadian tribunal (British Columbia’s Civil Resolution Tribunal) ruled that Air Canada was liable for information provided by its own chatbot, which had given a customer incorrect details regarding a bereavement refund policy. The company argued that the chatbot was a separate entity. However, the tribunal rejected this argument and set a legal precedent: companies are responsible for the information provided by their AI systems, regardless of its accuracy. Delegating communication authority to AI systems prone to "hallucinations" (a term used in the field to describe the generation of false factual information or conclusions) is a recipe for operational and reputational disaster.

Various studies have revealed significant rates of factual errors in legal responses. For instance, a 2024 Stanford University study showed that AI models specializing in law produce serious factual errors. Lawyers in the US have been caught citing fictitious legal cases—such as *Mata v. Avianca* (2023)—generated by AI in actual court filings. This case illustrates the fundamental failure of relying on an unverified and uncertified system that cannot distinguish between reality and plausible fiction.

Porcha Woodruff, a pregnant woman from Detroit, was arrested for armed robbery based on a misidentification by an AI facial recognition system. Police treated the algorithm's output as evidence rather than a lead requiring human verification. The case received media coverage and sparked heated debates regarding the use of AI in law enforcement, highlighting the traumatic consequences for innocent individuals.

Such failures reveal a fundamental issue: solutions—empowered to act—that lack conscience, accountability, and clear risk management systems. It is precisely in the gap between AI’s impressive computing power and corporate responsibility that the need for rigorous international standards arises—not to stifle innovation, but to limit risks and errors.

ISO/IEC 42119 – The Future of AI System Testing

The international standardization community's latest response to the challenges described above takes the form of the ISO/IEC 42119 series—an innovative set of international standards dedicated to testing artificial intelligence systems. The series is designed to serve as a cornerstone for organizations aiming to develop and deploy safe, robust, transparent, and trustworthy AI systems.

This initiative begins with the publication of the ISO/IEC TS 42119-2:2025 standard—an overview of AI system testing. It provides a comprehensive introduction to AI-related testing concepts, processes, and approaches, demonstrating how established ISO/IEC/IEEE 29119 software testing standards can be leveraged within the context of artificial intelligence.

The document covers the AI ​​system lifecycle and risk-based testing; testing processes and documentation tailored to AI specifics; testing levels and types—including data and model quality; guidelines for identifying and addressing risks specific to AI systems; and AI-specific test design techniques and coverage metrics. The ISO/IEC 42119 series is currently under active development, with new specialized components in the works:

The ISO/IEC 42119 series is designed to complement ISO/IEC 42001. While ISO/IEC 42001 establishes requirements for responsible AI governance, ISO/IEC 42119 provides the technical foundation for system testing and validation, helping to ensure reliability, safety, and trust in their use.

ISO/IEC 42001 — The benchmark standard for AI governance

Artificial intelligence has long since moved beyond the experimental phase, becoming critical infrastructure for companies, public administrations, and operators of essential services. The need for an internationally recognized framework for AI governance is evident. The answer has come in the form of ISO/IEC 42001—the first international standard dedicated to Artificial Intelligence Management Systems.

Why is ISO/IEC 42001 necessary?

The accelerated adoption of AI in business brings significant operational benefits but also tangible risks: algorithmic errors, a lack of transparency, hallucinations, security issues, or legal non-compliance. At the same time, international regulations—including the EU AI Act—are increasing pressure on organizations to demonstrate control, accountability, and traceability.

ISO/IEC 42001 provides a structured model for identifying and assessing risks associated with AI systems, establishing clear governance policies and controls, monitoring performance and ethical impact, ensuring compliance with legal and regulatory requirements, and integrating with other existing management systems (such as ISO 9001, ISO 27001, etc.). The standard follows the logic of established management systems, facilitating its integration into an organization's existing integrated management framework; this transforms it from a mere compliance exercise into a tool for genuine organizational maturity.

The new legislation on artificial intelligence emphasizes accountability, risk assessment, and user protection. ISO/IEC 42001 certification thus becomes a strategic tool for organizations that develop, implement, or use AI solutions. Through certification, companies demonstrate that:

• Manage AI in a controlled and documented manner
• Mitigate reputational and legal risks
• Offer partners and clients a high level of trust and transparency
• Are prepared for compliance requirements under the EU AI Act and other emerging regulations

In sectors such as aviation, critical infrastructure, healthcare, legal services, or financial services, this framework can represent the difference between formal compliance and genuine leadership in the governance of new technologies.

Global milestones: organizations that have already chosen the path to certification with SGS

The global adoption of ISO/IEC 42001 confirms that it is not merely a theoretical tool but a practical framework applicable across various sectors. The SGS Group has played a pivotal role in certifying pioneering organizations worldwide.

A defining moment was the certification of Changi Airport Group—the operator of Singapore Changi Airport—marking the first accredited ISO/IEC 42001 certification globally. In an airport ecosystem where AI supports critical processes, ranging from security to passenger flow management, the implementation of a formal AI management system sends a powerful message regarding control, accountability, and operational safety.

In Europe, AI Clearing and Xayn were among the pioneers of certification, demonstrating the standard's applicability to both AI-assisted infrastructure and construction and the development of software solutions. In the United Kingdom, Brookcourt Solutions solidified the standard's position within the cybersecurity sector.

In Asia, the adoption process was equally dynamic: OrionStar Robotics secured the first ISO/IEC 42001 certification in China; Japan’s Godot Inc. marked a similar milestone for the local market; DYXnet demonstrated the integration of AI governance in telecommunications; and KPMG India validated the standard's applicability within the professional services sector.

These certifications, awarded across diverse sectors—such as aviation, robotics, cybersecurity, and consulting—confirm that ISO/IEC 42001 is a practical governance tool applicable to any organization using AI responsibly. These milestones give rise to a new paradigm: in the AI ​​era, leadership is defined not only by the capacity to innovate but also by the ability to manage the technology's impact and the risks associated with its use in a responsible and transparent manner.

How SGS supports its partners on this journey

SGS, a global leader in testing, inspection, and certification, offers an integrated service portfolio to help organizations navigate the dynamic landscape of artificial intelligence standards, regulations, and best practices.

The company provides ISO/IEC 42001 accredited certification for AI management systems, alongside other internationally recognized certification schemes. The goal is not merely superficial compliance, but the establishment of a genuine, sustainable governance system that is auditable and credible to all stakeholders. AI system testing encompasses the evaluation of data quality, fairness, and transparency, as well as adversarial testing. SGS also conducts technical product testing for LLMs, chatbots, and Generative AI systems—including red teaming and security assessments—addressing the very risk categories illustrated in the real-world cases presented earlier.

SGS’s expertise covers AI risk management and preparation for compliance with international standards and regulations. SGS Academy offers courses dedicated to AI standards, the EU AI Act, risk management, and relevant technical topics—enabling organizations to be not only compliant but also competent. For organizations wishing to quickly assess the maturity of their AI governance, SGS AI Trust Check provides a rapid self-assessment tool that helps identify gaps and prioritize the actions needed for compliance and trust-building. Choosing an AI assurance partner should not be a spur-of-the-moment decision; it is a strategic decision with long-term implications. SGS brings the following to the table:

• over 500 digital trust experts, 10 accredited cyber laboratories, and a presence in more than 100 countries
• the first provider of accredited certification for AI management systems and comprehensive EU AI Act compliance services
• integrated services covering certification, testing, training, consulting, and digital trust solutions for every stage of the AI ​​journey
• extensive experience in financial services, healthcare, mobility, manufacturing, aviation, and more
• operational excellence and Swiss-standard rigor, dedicated to enhancing your organization's credibility

The difference between an organization that merely uses AI and one where the AI ​​system operates safely, responsibly, and reliably is measurable, auditable, and certifiable. Contact the SGS team to learn how we can support you in leveraging the upcoming ISO/IEC 42119 series and implementing ISO/IEC 42001 now.

About SGS

SGS is the world’s leading testing, inspection, and certification company. We operate an extensive network of over 2,500 laboratories and operational units across 115 countries, supported by a team of more than 100,000 dedicated professionals. With over 145 years of experience in operational excellence, we combine Swiss precision and rigor to help organizations achieve the highest levels of quality, compliance, and sustainability.

Our brand promise—"when you need to be sure"—reflects our unwavering commitment to credibility, integrity, and reliability, providing companies with the assurance they need to perform and grow with confidence. We deliver specialized services under the SGS brand, as well as through a portfolio of trusted brands such as Applied Technical Services, Brightsight, Bluesign, and Nutrasource.

SGS is listed on the SIX Swiss Exchange under the ticker symbol SGSN (ISIN CH1256740924, Reuters SGSN.S, Bloomberg SGSN SW).