Why we can’t ignore AI governance

What are the challenges?

As companies race toward artificial general intelligence — often defined as highly autonomous systems that outperform humans at the most economically valuable work — regulation is struggling to keep pace. The EU AI Act was first proposed by the European Commission in April 2021 but didn’t enter into full force until August 2026. This regulatory time lag leaves a critical governance gap. In our view, until regulation catches up, companies developing and deploying frontier AI must demonstrate that they can self-regulate in a secure, responsible and ethical manner.

That requires more than principles, ethics statements or voluntary commitments. It requires strong, expert and empowered governance: boards and executives with the authority, expertise and incentives to assess the impacts of AI on people, democracy, security and the environment — and to slow down, constrain or stop deployment where risks cannot be responsibly managed.

We are a long way from achieving this. Indeed, the trend seems to be in the opposite direction. According to World Benchmarking Alliance (WBA) Ethical AI Collective Impact Coalition (CIC), in 2024, 38 out of 71 companies (53%) explicitly embedded human rights into their AI principles, compared to 31 out of 52 (60%) in 2023.[2]

Is AI governance an investment issue?

Absolutely! AI provides compelling investments opportunities which are only matched by the high risks they incur. I’m thinking of human rights, legal and regulatory risks, operational risks, reputational risks etc. It is absolutely crucial that companies equip themselves as soon as possible with the strong governance mechanisms to manage AI risk effectively. This mean, for example:

• A strong responsible AI policy;
• An AI oversight committee composed of experts, both internal and external;
• The means and mandate to impact decisions on existing and new AI development and deployment;
• Accountability at the highest level of management and board;
• Staff Training.

Strong AI governance is the most important expectation of our Ethical AI investor coalition.

How does Candriam analyse these risks?

Our ESG Research team has developed its own AI framework. The framework is based on the level of risk associated with the various AI-use cases. It checks that companies have put the right policies, governance, risk identification and due diligence processes in place.

We use the EU ‘Double Materiality’ framework which looks at what sustainability risks a company faces, and how the use of AI by that company affects the world:

• ‘Outside-in’ risks include algorithmic biases which can create reputational and legal damage, and cybersecurity risks.
• ‘Inside-out’ risks include greenhouse gas emissions, discrimination from biased algorithms, and amplification of incorrect data. AI already consumes up to half of data centre power and as much as 3% of global electricity.[3] These data centres use four times as much water as previously thought.[4]

How do you engage with companies on AI?

The tensions mirror those of other engagement topics, but are more intense, and the financial stakes are potentially higher. High, and rising, private ownership of AI developers means public disclosure is limited to a smaller portion of the industry. Alongside this, there is inconsistent regulation across nations, even within the EU.

These contribute to a structural mismatch between the risks and our leverage as investors:

• We have limited access to the key actors: The leading AI developers are either privately held or unresponsive. Several of the leaders in this space have dual share class systems, meaning that investor voices are limited.
• There is an enormous concentration of power: A small number of hyperscalers controls the infrastructure and the models.
• The result of this mismatch is that leverage is indirect and uneven. Therefore, all our portfolios can be impacted from AI risks.

With AI engagement the devil is in the (lack of) data. Not just lack of data, but lack of comparability. This mirrors the early stages of other investing and engagement topics but with AI, we have less access and less leverage. However, we believe that the experience we have had addressing other topics could also be used to deal with the challenges related to AI. 

What else are Candriam and other asset managers doing to address these governance issues?

We are one of four co-leaders[5] of the WBA Ethical AI CIC, a group of more than 60 investors with over $11 trillion assets under management. [6] We aim to address the risks linked to the societal effects of AI by encouraging companies to commit to public guidelines on AI, to adopt a strong and expert governance, and to implement processes to ensure the safe development, deployment and use of AI.

Importantly, we were early to the topic: We began in 2020 with facial recognition and human rights. Through this collaboration with other investors, and our membership on the WBA steering committee, we have continued to raise awareness, co-leading engagements with selected companies.

What progress have you made so far?

Some progress has been made – the risks are increasingly well-understood, and governance frameworks are emerging, albeit among a minority of companies.

Owners are beginning to act. In the US, even while shareholder resolutions on social topics declined overall in 2025, resolutions for data security, privacy, and AI remained at a high level.[7]

But we can’t let our guard down. Of the WBA’s 200-company index, fewer than 40% have a publicly stated AI policy.[8] Only 12% disclose their actual governance mechanisms.[9] This lack of transparency makes it challenging to gauge if the policy is being implemented in the actual business. This also means it is difficult to monitor and share best practices.

The pace is also slowing: 19 new companies published AI principles in 2024, but only nine new companies unveiled AI policies in 2025.[10] There is limited implementation of guardrails at any scale, modest evidence to date of real-world impact, and persistent blind spots among AI developers and on the part of governments.

Recognition