By Annie Chong | Independent Technology & Transformation Advisor | SkillsFuture Fellow 2025 | PhD Candidate, Organisational Leadership | Board Director & Founder, ACMP Singapore Chapter | Her Courage Award 2026 Recipient
The real challenge in AI is not only technical capability. It is whether the people shaping these systems can see what the systems themselves miss. That is why the conversation about women in AI cannot stop at representation. Across Asia Pacific, women hold roughly 30 percent of AI roles, designing, training, and governing the fraud-detection and customer-safety systems that millions rely on, yet their expertise is still too often questioned when bias risks surface. This is not coincidence. It is a pattern. And when that pattern gets encoded into AI, it stops being a human failing and becomes infrastructure, quietly and systematically shaping who advances and who does not, at a scale no individual bias ever reached.
For more than three decades in global technology leadership, I have walked into rooms that were not expecting me and stayed anyway. In that time, I have seen how credibility is unevenly granted, how leadership is narrowly defined, and how systems quietly reflect those realities. What is changing now is not the existence of bias, but the scale at which it operates. Artificial intelligence is taking human assumptions and turning them into systems that can repeat those assumptions faster, wider, and with greater confidence than any person ever could.
That is why this conversation is not really about technology alone. It is about power. AI is becoming infrastructure for opportunity: who gets hired, promoted, trusted, flagged, or overlooked. In that world, who sits at the table when these systems are built, and what capabilities they bring, is not a symbolic question. It is a question with consequences that compound with every deployment.
The data makes the stakes concrete. A Russell Reynolds analysis of 39 leading AI organisations globally found that four had senior leadership teams comprised entirely of men, and 30 had less than one-third women. According to the Center for Equity, Gender and Leadership’s 2021 playbook on mitigating bias in AI, a Berkeley Haas analysis of 133 AI systems across different industries found that 44 percent showed measurable gender bias in testing. Amazon scrapped its own AI recruiting tool after it systematically downgraded applications containing the word ‘women’.
The pattern holds across independent sources. Stanford researchers found that large language models portray women as younger and less experienced than men, even when given identical information. The ILO’s March 2026 research brief confirms that female-dominated occupations are almost twice as likely to be exposed to generative AI displacement as male-dominated ones. Each of these findings points to the same conclusion: AI is not just reflecting bias. It is institutionalising it, turning subjective historical judgments into what appear to be objective scores, and scaling them across millions of decisions.
“We did not call them soft skills because they lacked strength. We called them soft because we valued what was easy to measure over what was hard to see and then wondered why women kept being told they were not technical enough. Now AI is building those instruments. And we are paying the debt for that oversight.”
Let me give you one small example of how this plays out. Not in a boardroom, not in a governance report, but in an ordinary working moment. I once used AI to develop a strategic plan. A senior male colleague reviewed it, assumed I had not used AI, and proceeded to explain to me how AI could improve my work, entirely unaware that the document in front of him was already AI assisted. He was using AI and was presumed capable. I was using AI and was presumed not to be. That gap in assumption is not unique to me. It is exactly what we are now encoding into the systems that will assess the next generation of women leaders.
In regulated industries, where I have spent much of my career, we operate from a foundational principle: a risk that is unacknowledged cannot be managed. That is exactly the problem with biased AI. It is often invisible because it arrives dressed as objectivity, and it is accelerating because adoption is moving faster than the oversight needed to contain it.
But I want to make a different argument from the one most commonly heard in this space. The conversation about women and AI tends to focus on representation: not enough women in these roles. That is true, and it matters. What is far less discussed is what organisations actually lose when the full range of human capability is not brought to bear on AI development. This is not a story about women versus men. It is a story about what happens when organisations define capability too narrowly and consequently build technology that is less reliable than it could be.
In my own work, I developed what I call the Decision-Delegation-Accountability (DDA) Framework™, a practitioner framework built from three decades of field observation in regulated industries. It identifies three layers. Layer 1: decisions that must always stay human, including ethical judgments, strategic choices, and people decisions requiring values and lived experience. Layer 2: AI recommends and humans approve, with full accountability retained at every step. Layer 3: decisions fully delegated to AI, low-risk, repetitive and rule-based, within explicitly defined parameters. Scope creep beyond this layer is a governance failure. The overarching principle across all three layers: accountability must always stay with humans. As I have put it directly, this is not a limitation of AI. It is a requirement of leadership. The capabilities that make this framework work are not the ones we have historically rewarded. They are the ability to ask who this system affects and how, and the courage to challenge an output that looks confident but may be encoding assumptions nobody has examined. We have spent decades calling these skills soft. AI governance is now revealing them to be system-critical.
To make this concrete: in a telecom fraud detection system, flagging a suspicious transaction sits in Layer 2, AI surfaces the pattern, but a human approves any action affecting a customer. Permanently blocking an account sits in Layer 1. Routing routine queries sits in Layer 3. Accountability across all three layers stays with a named human. That is the governance logic that distinguishes responsible AI from AI that produces liability. Yet for too long, the measure applied to AI capability was borrowed from a previous era: engineering depth, infrastructure command, the ability to build the model. That standard made sense when technology was about systems. It makes far less sense when technology is about decisions, and the decisions AI now makes affect careers, access and opportunity. McKinsey’s 2026 AI Trust Maturity Survey found that strategic oversight and agentic AI controls lag behind technical capabilities in organisations across every region globally, with technology, media and telecommunications among the sectors facing the sharpest governance gaps. ISACA confirmed this year that boards are increasingly being asked to treat AI not as a technical topic but as a strategic and risk issue. The question that matters is no longer whether the system works. It is whether it can be explained, challenged, corrected and held to account. These are not engineering questions. They are leadership questions. And the capabilities required to answer them, namely judgment, ethical reasoning and the courage to challenge outputs nobody else is questioning, are precisely what has been systematically undervalued, because they were being measured against the wrong standard.
The World Economic Forum’s Future of Jobs Report 2025 identifies leadership and social influence, resilience and adaptability, and creative thinking among the fastest-rising skills demanded by employers through 2030. LinkedIn data from the WEF Global Gender Gap Report 2024 found that women hold a 28 percent larger share than men of precisely these capabilities. Research from Russell Reynolds Associates found that women leaders are more likely to advocate for ethical AI oversight and to raise concerns about bias risks. The point is not that these capabilities belong to one gender. The point is that all of them are needed, and diverse teams that bring both technical and human-centred strengths together are more likely to build AI that actually works for the people it is meant to serve.
There is also a technical dimension to this that is rarely named. As AI moves deeper into unstructured data, including language, behaviour, sentiment and context, the gap between what the model produces and what it actually means widens. Interpreting that gap requires precisely the capabilities we dismissed as soft: the ability to read what is not said, to sense what the data does not capture, to know when a confident output is missing something human. The frontier of AI is not just pattern recognition in clean datasets. It is meaning-making in messy, ambiguous, deeply human territory. Women who have spent careers navigating rooms where the explicit and the implicit rarely matched have been developing exactly that capability. It was never soft. It was always the hardest problem.
There is also a governance risk that organisations rarely name until it is too late. I call it the Assist-Rely-Defer progression: organisations begin by using AI to assist human decisions, then rely on it, and eventually defer to it entirely, accepting outputs without interrogation. I have sat in rooms where AI-generated assessments sailed through leadership review without a single question, not because the leaders were incapable, but because the output looked authoritative and nobody felt it was their place to challenge it. The shift is not AI becoming smarter. It is humans becoming less critical. Women who have spent decades questioning assessments that undervalued them and challenging definitions that excluded them carry a practised scepticism that is exactly what responsible AI oversight requires. That scepticism is not incidental. It is a structural safeguard.
Trust is at the centre of this year’s theme, and it deserves honest examination. Real trust in AI cannot be passive. It must be earned through transparency about how systems work, through accountability when they cause harm, and through the genuine inclusion of those most affected in shaping them. Women across the region are being asked to trust systems built largely without them, trained on data that did not fully represent them, and overseen by frameworks that rarely include them. That is not trust. It is compliance. And this matters especially across Asia Pacific, where telecom operators and digital platforms are shaping the systems that increasingly govern work, safety and opportunity. AI here will not be judged by how fast it is adopted. It will be judged by whether people can trust it to serve them fairly.
—
What needs to change is both structural and cultural. And it needs to change now, not because this is a future risk but because the systems making these decisions are already running. For telecom operators and digital platform builders specifically, who sit at the intersection of AI deployment, network safety, and the digital lives of hundreds of millions of people across this region, the question of who governs AI outputs is not a future consideration. It is today’s operational risk. AI systems that influence who get selected, advanced or identified as capable must be subject to mandatory gender bias audits, conducted by diverse teams and acted upon. Women’s AI fluency must be treated as a strategic priority: the WEF Future of Jobs Report 2025 found that nearly 40 percent of core skills will change by 2030, and women excluded from shaping that shift will bear the highest cost. Culturally, the definition of what strong AI leadership looks like must expand. The best teams will not be those with the deepest technical bench alone. They will be those that combine technical expertise with the human judgment to question what the model cannot question for itself. That is what AI that can genuinely be trusted looks like.
This is also what drives my own work. Through the ACMP Singapore Chapter, the Odyssey Mentoring Programme, and doctoral research in Organisational Leadership, I am building the infrastructure that the people not yet in these rooms will need when they arrive. The question of who shapes AI is the question of who shapes what comes next — and right now, that question still has an uncertain answer.
I think about the women across Asia Pacific building AI systems right now, carrying the weight of having to prove what their male peers are assumed to know, doing extraordinary work in rooms that do not always fully see them. I think about the women I mentor, capable and prepared, who will be assessed by systems that may quietly undervalue them because the data those systems were trained on did not fully account for people like them. And I think about what every organisation is losing when complementary strengths are not brought together. That loss is not only a gender equity problem. It is a reliability problem.
The question is not whether AI will continue to advance. It will. The question is whether the people shaping it will have the full range of vision, capability and judgment that earning trust actually demands. The answer depends on whether organisations are willing to ask themselves honestly: when did someone last challenge an AI output in your organisation, and what happened when they did? When you picture the person leading your AI decisions, what does that person look like, and what capabilities do they have? And if the answer to that second question does not trouble you, perhaps it should. The technology will reflect whoever shapes it. Make sure women are in that room.
Those are not soft questions. They are the questions that determine whether AI in Asia Pacific becomes infrastructure that works for everyone, or infrastructure that quietly works against the very people who helped build it.
Annie Chong is a Global Technology Executive, Independent Director and Practitioner-Scholar with over 30 years of experience leading enterprise-scale digital transformation across Pharma, Biotech and Advanced Manufacturing in Asia Pacific. She is Founder and Vice President of the ACMP Singapore Chapter, Board Member of the Cloud Security Alliance Singapore Chapter, Member of the Singapore Institute of Directors, and a doctoral candidate in Organisational Leadership at Columbia International University. She is a recipient of Singapore’s SkillsFuture Fellowship 2025, the nation’s highest individual skills recognition, conferred by the President of Singapore. She mentors women in technology leadership and advocates for AI governance, responsible transformation and human-centred change. She is an awardee of the Her Courage Award 2026.
Sources
International Labour Organisation, March 2026. Gen AI, Occupational Segregation and Gender
Equality in the World of Work. ilo.org
Russell Reynolds Associates, Q4 2024. AI for All: The Case for More Women Leaders in AI.
russellreynolds.com
Center for Equity, Gender and Leadership, Berkeley Haas, 2021. Mitigating Bias in Artificial
Intelligence (analysis of 133 AI systems).
Reuters, October 2018. Amazon scraps AI recruiting tool that showed bias against women.
Guilbeault, D., Delecourt, S. and Desikan, B.S., 2025. Gendered age bias in artificial intelligence and
online images. Nature Human Behaviour. Stanford Graduate School of Business.
https://doi.org/10.1038/s41562-025-02131-3
World Economic Forum, 2025. Future of Jobs Report 2025. weforum.org
World Economic Forum, 2024. Global Gender Gap Report 2024, LinkedIn Economic Graph data.
weforum.org
Russell Reynolds Associates, H1 2024. Global Leadership Monitor. russellreynolds.com
McKinsey, March 2026. State of AI Trust in 2026: Shifting to the Agentic Era. mckinsey.com
ISACA, February 2026. Responsible AI: From Emerging Technology to Executive Governance
Imperative. isaca.org