Alvin Toffler correctly pointed out that the illiterate of the 21st century are those who can’t learn, unlearn and relearn, not those who can’t read and write – and his words are all the more relevant in today’s artificial intelligence (AI)-shaped world.
Yes, we’ve become quite comfortable using AI, but do we fully trust it? According to a wide-ranging global 2025 collaborative study between KPMG and the University of Melbourne, more than 48,000 people across 47 countries were asked about AI and its impact on enterprises and individuals and organisations.
The objective? To know about the public’s attitude, use, and trust towards AI. It showed that while 90% of people recognise AI as a valuable tool, and while the number shows more optimism that scepticism, only 76% trusted its outputs. This gap is rooted in some very real concerns: insufficient oversight, potential misuse, and safety risks, and ends up impacting real-world outcomes – because what stands between humans and the complete promise of AI is trust.
The Threat of Shadow AI
One of the most immediate threats when we talk about the lack of trust in AI isn’t sophisticated external attacks. Rather, it’s the holes and gaps in security created by well-intentioned employees who are trying to work better and more efficiently. We’re talking about “shadow AI,” where employees using public AI tools not sanctioned by the enterprise have been creating huge, unmonitored security holes across organisations.
One of the most powerful cautionary tales in recent times is the May 2023 accidental data leaks in Samsung. In three separate cases, a group of well-meaning engineers unintentionally ended up leaking highly sensitive corporate data, all because they used public AI tools for work assistance.
The AI tool in question? ChatGPT, a tool that we all use. The engineers entered faulty source code in the AI tools looking for solutions, entered proprietary codes for optimisation, and ended up uploading confidential meeting notes for generating minutes.
This was a failure on part of the enterprise, not the employees, to provide secure effective tools for use. And when teams need to decide between security and efficiency, organisations have already lost the plot, and this behaviour is already widespread in use. Even though enterprises have gotten better at tracking employee behaviour when it comes to using such AI tools, employees are still using shadow AI, according to research from AI security company BlackFog.
According to the study, nearly half of workers still use AI in ways that their bosses don’t approve of. Half of this number was using free versions of software, so there was no enterprise-level security and/or data governance. Unsurprisingly, the consequences of such wanton use are bound to be substantial.
Privacy And Security – Top Concerns For Consumers
No conversation about trust deficit in AI at the consumer level is complete without addressing agentic AI and its security – or the lack of it. Developers and customers have the same end goal: they want agentic AI to be protected. However, the majority isn’t convinced that AI agents will respect their privacy and security needs, according to the 2025 Auth0 Customer Identity Trends Report. 60% respondents said they were extremely concerned about AI’s impact on their digital security and privacy.
There’s no doubt that education and awareness can bridge this gap, at least partially. 44% of those who didn’t use AI said that they didn’t trust agentic with their personal data. So, disclosing what sensitive data agentic AI is accessing explicitly and how it’s being used could be a powerful step toward building customer and developer confidence.
The Humans + Machines factor that keeps humans in the loop might also help, even though it trades AI bias for human bias and risks introducing human errors and delays. When we look at the advantages, it could possibly train agents to understand the nuances of a request better and refine their responses. Meanwhile, developers could take steps to design and deploy more secure AI apps.
Identity Security – The Key To Navigating The AI Era
When it comes to privacy and security, a majority of it hinges on getting identity security right. Agents with more than necessary permissions could result in not only data leaks but also increase malicious activity possibly. Imagine AI agents who suggest products for your shopping list have access to your medical records when you inquire about an OTC (over-the-counter) medication. Now, imagine that very AI agent falling victim to prompt injection attacks – the security vulnerabilities are serious.
Essentially, the power that AI agents possess requires changing the attack surface that organisations and enterprises need to protect. Furthermore, it’s important that businesses and institutions employ the concept of least privilege and ensure that dynamic updates keep taking place for permissions to revoke or grant authorisation when needed. In the end, it all boils down to zero trust.
While there’s still a lot of work to be done, there’s every reason for optimism. In the 2025 Auth0 Customer Identity Trends Report, of those who preferred AI agents over humans, a little over half said that they view AI technology as the future.
Agentic AI and AI tools do offer a path to efficiency, yes, but in order to walk it securely and efficiently, business leaders need to demonstrate to consumers that they can offer a user experience that’s not only reliable and fast, but also, most importantly, secure.
In case you missed:
- How Zero Trust Works in the Agentic AI Era
- Hiding In The Dark: Navigating The Threat Of Shadow AI
- Agentic AI and its Future in the Fintech Revolution
- AgentOps: The Dawn Of The Internet Of Agents
- The Multi-Cloud Security Blueprint for the AI Era
- The Rise Of Agentic Cloud – Reshaping Cloud Infrastructure In 2026
- AI Runtime Security: A Comprehensive Look
- Are Agentic AI Browsers Safe?
- Agentic Commerce: A New Era Of AI-Assisted Retail
- Blockchain-Based Digital Identity: The Future Of Verification Is Here
