Responsible AI: Building trust and long-term value in the age of algorithms

We are continuously seeing how developments in AI are shaping the market, whether it's resulting in job redundancies, child safety discussions, or national security. AI is scaling faster than the guardrails around it, showcasing how Responsible AI has become a material issue for companies and investors. In this paper, our Senior ESG analyst Rita Wyshelesky highlights our research approach, our proprietary Responsible AI due diligence framework and engagement priorities.

Artificial intelligence (“AI”) has rapidly evolved from a niche technology to a central driver of corporate transformation, accelerated by the rise of Generative AI (“GenAI”). Its impact spans sectors; from financial firms using predictive analytics to optimise portfolios¹, to healthcare systems applying machine learning for early disease detection². Yet the same technologies that create value also introduce risks, including bias, misinformation, and systemic disruption.

The 2025 Stanford AI Index reports a 56.4% increase in AI-related incidents in 2024³, and deepfake content is projected to surge from 500,000 files in 2023 to more than 8 million in 2025⁴. With reportedly 71% of companies now regularly using GenAI⁵, these trends underscore the urgency of adopting Responsible AI frameworks that ensure innovation is aligned with safety, trust, and long-term value creation.

The term “Responsible AI” (or sometimes “Ethical AI”) is widely used yet variably interpreted. At its core, it refers to the principled design, deployment and oversight of AI systems in ways that respect human rights, ensure safety and build trust. This concept is key to ensuring that technology scales in a measured way that avoids societal damage or attracts over-regulation.

While existing frameworks differ, five pillars appear consistently across regulatory and industry standards^6,7:

• Fairness - AI systems should not systematically disadvantage individuals or groups (e.g. gender, race, age, or disability).
• Explainability – Stakeholders should be able to understand, at the right level of detail, how an AI system arrives at its outputs.
• Privacy – Systems should have robust data protection and constraints on secondary use of data and protection of individual’s human rights.
• Robustness, Security, and Safety - Mechanisms should be in place to manage risks, prevent undue harm, and allow systems to be overridden or decommissioned.
• Accountability - Clear human accountability must sit above automated systems. Responsibilities for AI risk management, model validation and incident response need to be assigned at management and board level, and documented in policies, controls and escalation procedures.

These values are integral to minimising operational and reputational risk. Poorly governed AI can result in biased credit decisions, unsafe medical recommendations or flawed trading signals, translating into an increase in legal exposure, remediation costs and regulatory penalties for companies.

From a reputational point of view, AI incidents can undermine customer trust and brand equity far quicker than traditional operational failures. For example, Google’s Bard demonstration error in 2023 wiped roughly $100bn from Alphabet’s market value within days, underscoring how AI hallucinations can trigger outsized reputational consequences⁸. Microsoft also suffered from similar issues, when Bing’s chatbot was sending users threatening or hostile messages back in 2023, leading to a share price drop of 4% in response to the events⁹.

Although the potential negative consequences of AI are considerable, current regulatory measures have not kept pace with rapid technological progress.

The EU AI Act¹⁰, which stands out as one of the most comprehensive frameworks for AI governance, remains largely reactive, emphasising risk classification and compliance over systemic accountability. And even that faces the possibility of dilution as a result of Big Tech lobbying¹¹.

The U.S., UK, and Asia-Pacific markets, on the other hand, have largely adopted principle-based or voluntary frameworks, leading to inconsistent implementation. This fragmented landscape creates uncertainty for global companies and encourages “jurisdiction shopping” for laxer environments.

While regulations set minimum compliance standards, stakeholders increasingly expect organisations to uphold higher ethical benchmarks. This shift has led to growing demands for greater transparency, algorithmic fairness, and responsible environmental management.

For investors, Responsible AI translates into three main workstreams:

1. Integration into ESG analysis and research: Responsible AI can be integrated into existing ESG frameworks, rather than treated as a separate overlay. Environmental considerations are increasingly relevant as well, particularly the energy intensity and carbon footprint associated with training and operating advanced AI models.

2. Enhanced due diligence on AI governance and model transparency: Investors can engage with companies to drive further transparency on AI governance and evaluate how the company monitors potential adverse impacts.

3. The rise of AI governance metrics and reporting standards. While still nascent, we can expect the emergence of industry-specific AI governance KPIs.

Carmignac incorporates material ESG factors into its investment analysis and decision-making processes, acknowledging their potential to significantly impact risk and return. Therefore, when analysing companies where AI is strategically relevant, we would typically look for:

• Clear strategy and materiality: Evidence that management understands where AI is value-creating and where it introduces material risks.
• Robust governance: Integration of AI oversight into existing governance structures, including having board literacy on AI, documented policies, and alignment with recognised frameworks.
• Data and model discipline: Strong data-governance practices, rigorous model validation, and transparency appropriate to the sensitivity of use-cases.
• Alignment with stakeholders and long-term value: Consideration of employees, customers and society.

These factors guide our stewardship activities (voting and engagement[LM1.1][RW1.2]), with the objective of enhancing risk-adjusted returns while promoting resilient, trustworthy AI adoption. In addition to individual engagement with companies, we also engage collaboratively with corporates through the World Benchmarking Alliance’s Collective Impact Coalition for Ethical AI. With regards to proxy voting, we review shareholder proposals on a case-by-case basis, and have historically supported shareholder resolutions with companies like Meta, Alphabet and Amazon, asking for increased transparency on GenAI risks.

Our Responsible AI due diligence framework translates principles into concrete scores that can be used to make and compare decisions. The framework assesses companies across seven pillars, from governance and risk management to fairness, transparency and environmental impact, using a structured, KPI-based checklist.

We have identified a lack of structured performance analysis for Responsible AI – one that investors can use to price this material issue. As a result, we have developed our own framework. These indicators are scored and normalised into an overall Responsible AI score, allowing us to distinguish leaders from laggards across AI developers, deployers and hybrid players.

We find that the highest scores are in our User Rights, Feedback & Redress metrics (excluding energy, which is covered by only one KPI) while the lowest scores are in Content Integrity. This outcome is expected given the current stage of AI deployment. While organisations are establishing appropriate governance structures and policies, these measures have not yet been fully implemented across all product suites or consistently reflected in outcomes.

When we overlay these scores with real valuation multiples¹², we do not find direct correlation across all categories, implying that the market is yet to take Responsible AI into account. This is to be expected given the lack of standardised information for the market to absorb. Whilst for AI infrastructure platforms we found that the top-scoring listed names in the framework traded on higher P/E multiples, this was not the case for other sectors within the AI supply chain. Currently, growth, margins, market position and general governance remain the main drivers of valuation, but we anticipate this will change in the future as the topic becomes increasingly material.

We believe that with time, companies that proactively manage AI ethics will trade at higher valuation multiples due to three reinforcing dynamics:

1. Reduced Risk Discount: Investors assign lower risk premiums to firms with strong governance. Transparent AI governance lowers exposure to litigation, regulatory fines, and brand crises.

2. Enhanced Innovation Capacity: Firms with mature Responsible AI frameworks often exhibit more disciplined data management and governance processes, which improve model efficiency and accelerate scalable innovation.

3. Talent and Brand Premium: Evidence shows that strong ethical governance improves employer attractiveness and customer trust¹³. Therefore, companies with visible Responsible-AI commitments illustrate how such governance can strengthen these competitive advantages.

Beyond internal controls, the most material tests of Responsible AI are playing out in high-stakes societal domains where harms scale faster than safeguards:

1. Child safety in the age of GenAI

Child online safety has become an increasingly important topic of discussion, as the rapid evolution of technology makes it difficult to introduce appropriate controls in a timely manner. It is increasingly being shaped not only by governments but also by civil society and investor stewardship. We are increasingly seeing countries moving to tighten protections for minors online, including minimum age requirements and stronger age-verification expectations for social media platforms¹⁵.

Investors, meanwhile, are increasingly framing child safety as a governance issue; for example, shareholder proposals at companies such as Meta calling for clearer oversight, quantitative metrics, and greater transparency received 59% support among non‑management‑controlled votes in 2024¹⁶. At the same time, GenAI is emerging as a distinct new catalyst for action beyond social media.

Child-rights groups warn that GenAI can scale new risks¹⁷, such as synthetic sexual abuse material and more persuasive misinformation, and are issuing guidance aimed directly at AI developers and governments, not only content-hosting platforms. Policymakers are also beginning to scrutinise AI chatbots and companion-style tools¹⁸, signalling a shift in focus from “social media safety” toward broader “youth safety in AI-mediated digital services.”

2. Will AI cause mass unemployment?

One of the most sensitive aspects of GenAI is its potential impact on jobs. For many companies, large-scale automation can unlock productivity and margin expansion. But abrupt headcount reductions, opaque restructuring decisions or the use of AI in hiring without proper safeguards could trigger employee backlash, union disputes and reputational damage for companies.

While media narratives often link GenAI to widespread unemployment, current macroeconomic data does not indicate a measurable increase in unemployment attributable to AI¹⁹. The World Economic Forum projects that by 2030 AI will help create 170 million jobs, while displacing 92 million positions²⁰, indicating a net increase in vacancies. Historically, the diffusion of horizontal technologies has tended to reshape roles and reallocate work rather than reduce overall employment levels, though task restructuring is already visible in several industries, and the long-term distributional impact remains uncertain.

Radiology offers a useful illustration. In 2016, AI was predicted to replace radiologists within five years²¹. A decade later, the reality is quite different: radiologist numbers have increased, yet many healthcare systems still face shortages because imaging demand and clinical complexity have outpaced capacity²². Rather than replacing radiologists, AI has become a valuable complement, improving triage, quality, and workflow, while keeping expert oversight and accountability firmly in the loop.

Therefore, as investors we look for evidence on how companies manage AI’s impact on employment beyond short-term cost savings. Firms that treat AI as a complement to human capital, rather than a blunt cost-cutting tool, are more likely to preserve culture, attract talent and avoid social, regulatory or political pushback.

The question is no longer whether AI will reshape industries, but how responsibly that transformation will occur. In the event that large-scale unemployment emerges without effective economic reallocation, the introduction of an AI tax would become increasingly inevitable. Investors should now prioritise discussions on AI governance, transparency and long-term resilience, ensuring that innovation aligns with sustainable value creation.