Published 15 May 2025 in Artificial Intelligence • 6 min read • 
Audio available
From bias and copyright infringement to hallucinations, deepfakes, and false information, we are all familiar with the unhelpful (at best) and damaging (at worst) idiosyncrasies of AI that have overshadowed the development and adoption of one of history’s most significant technological advances.
For example, during the 2024 UK election, a video of former Prime Minister Rishi Sunak declaring plans to introduce compulsory military service in conflict zones appeared on social media. It was a deepfake; there were no such plans. And in 2018, Amazon was forced to scrap an AI-automated hiring tool after it discriminated against women due to male-oriented bias in its data.
Rogue AI can also have serious commercial consequences. In 2012, the financial services firm Knight Capital lost nearly half a billion dollars in less than an hour after an AI-powered algorithm triggered unintended stock trades. As a result, the company had to be sold to a rival. Unforeseen problems may well emerge with the rapid development of AI agents – systems based on Large Language Models (LLMs) that can operate autonomously on the internet and other platforms.
For organizations and leaders, these risks and realities represent more than just eye-grabbing headlines. They can make the difference between success and failure in a future that will likely be shaped by those who harness AI most effectively and credibly. Trust is at the heart of the matter. Society is already anxious about the existential repercussions of AI, such as mass job losses and concerns about autonomous models. Flaws in how AI operates add to these fears and threaten the basis of trust in technology to improve our lives.
What can businesses and executives do to encourage greater trust in AI – and their growing reliance on it – among their customers, policymakers, investors, employees, and wider society? Organizations and leaders must address two questions. Do you trust AI as a source of information for your work? And can you put your hand on your heart and say you trust the AI your organization is developing or using to deliver its services? For many of us, the answer is probably: “I can’t be certain.”
Given the vast number of cases where AI has failed the trust test and a lingering lack of faith in the technology, it’s crucial to have a robust “due diligence” process in place to minimize the risks. The nature and speed of AI’s advance may make this appear analog and cumbersome, but taking pains to adopt a responsible approach will likely save some embarrassment or much worse. Ultimately, it could differentiate your organization in the marketplace from others with weaker governance systems. Based on these two key questions, we’ve developed a practical checklist for you and your organization to strengthen faith in AI.
“Deepfakes are the fastest-growing social engineering vector in AI-enabled cyber-attacks. Using our eyes and ears is still the best way to differentiate between AI-generated and genuine human content.”
Always ask for and verify the sources or references behind GenAI output. In 2024, an AI-powered chatbot created by the New York City authorities to help small business owners made headlines when it provided
This had negative reputational implications for the city and risked business owners making potentially harmful and illegal decisions. Verifying the bot’s advice and sources would have limited the damage.
Maintain a critical mindset toward GenAI outputs and validate with secondary sources, especially for cybersecurity-related tasks. Deepfakes are the fastest-growing social engineering vector in AI-enabled cyber-attacks. Using our eyes and ears is still the best way to differentiate between AI-generated and genuine human content.
Look for user reviews and documented case studies that illustrate real-world performance and reliability. You can also tap into helpful monitoring tools such as the Galileo Hallucination Index, which measures the performance of leading LLMs.
Request clarity on the data sources and methodologies used to train AI to better assess potential biases. Most biased outputs can be traced to training data sets that were not carefully curated and were unrepresentative of the group for which the output would be used. Facial recognition systems are a great example. A study by Joy Buolamwini and her co-author Timnit Gebru (Proceedings of Machine Learning Research, 2018) showed the error rate for light-skinned men was 0.8%, but it was 34.7% for darker-skinned women. By asking for transparency in data sources, you will be able to control whether the training data set is fit for the purpose for which the AI is being trained.
“Take Duolingo: when it introduced its conversation practice tool Duolingo Max, which is powered by Chat-GPT, the language instruction company warned that AI-created responses may not always be accurate and encouraged learners to report errors.”
Erik Brynjolfsson of the Stanford Institute for Human-Centered AI has estimated that “ billions of dollars are being wasted” on AI by companies, with insufficient focus on generating value. Avoid investing in AI for its own sake; instead, focus on solving specific pain points where clear value can be demonstrated.
Transparency matters. Utilize model cards to document how your team assesses and mitigates risks (e.g., bias and explainability) and make this information available to users. Model cards accompany machine learning models to give guidance on how they are intended to be used, including performance assessment. For example, they can draw attention to how they perform across a range of demographic factors to indicate possible bias.
Implement regular audits of AI performance and fairness to proactively identify and address any issues. If you are unsure how to conduct this kind of audit, bodies such as the European Data Protection Board and the Dutch-based ICT Institute have published helpful checklists.
Define clear accountability structures and governance policies around your AI systems. This might include ethics boards or external audits to bolster user trust. For example, IBM has established an AI ethics board, while Fujitsu has set up an external advisory committee on AI ethics.
Educate your users about how AI works: its limitations and the measures taken to ensure its reliability. Clear, jargon-free communication can help demystify AI and build trust. Take Duolingo: when it introduced its conversation practice tool Duolingo Max, which is powered by Chat-GPT, the language instruction company warned that AI-created responses may not always be accurate and encouraged learners to report errors. So, a starting point could be as simple as informing a customer they are talking to a chatbot, not a human, and that there are limits to what it can offer help with – but that a person is on hand if needed.
While AI undoubtedly offers unprecedented possibilities for growth and operational improvements, organizations must be careful to balance the urgent need for trust against rushing to grab commercial gains.
Putting in place strong systems that continuously challenge the trustworthiness of the AI that you use and expect your customers to interact with marks an important first step in enhancing faith in the technology and the businesses that increasingly rely on it.
The independent International Organization for Standardization (ISO) has developed guidelines for managing risks around using artificial intelligence. This framework offers a helpful starting point for organizations looking to establish safer systems and processes to build trust in the fast-moving technology. Click here to find out more.
Companies are rushing to embrace the game-changing opportunities to improve operations and scale that AI offers; yet many are unaware of the pitfalls. Test your knowledge of the risks here – and...
In an era defined by advanced AI and behavioral data, Apple’s 'Evil Steve' test exemplifies a proactive approach to ethical decision-making and responsible data governance....
How a private equity-backed corporate carve-out created a successful, sustainable consulting powerhouse ...
It’s possible to immunize people against false or manipulated information through simple exercises that policymakers and platforms could easily roll out....
Audio availableExplore first person business intelligence from top minds curated for a global executive audience