AI is no longer a capability your organization is evaluating. It is already in your inboxes, your documents, your service desks, and your boardroom decks. The question is no longer whether your people are using AI. The question is whether they are using it well.
Enthusiasm is understandable. But a pattern is emerging across organizations of every size: confidence in AI is outpacing caution. That gap is where risk lives.
The Confidence Gap Is Real
The numbers are striking. According to the EY 2025 Work Reimagined Survey of 15,000 employees across 29 countries, 88% of employees use AI at work – but mostly for basic tasks like search and summarisation. At the same time, 37% of those same respondents worry that overreliance on AI could erode their skills and expertise.
Meanwhile, the Zscaler ThreatLabz 2026 AI Security Report, drawing on nearly 989 billion AI transactions across approximately 9,000 organizations, found a 93% year-over-year increase in employees transferring enterprise data to AI tools – totalling more than 18,000 terabytes. ChatGPT alone was linked to 410 million data loss prevention violations, including attempts to share source code, Social Security numbers, and medical records.
This is not a technology problem. It is a habits problem. And habits can be changed.
Start With the Prompt
The quality of your AI output is determined before the model ever responds – it is determined by what you put in. Prompt engineering is rapidly maturing from a niche skill into a core professional competency. Well-constructed prompts define a clear role for the AI, constrain the task scope, specify the desired output format, and include relevant context without oversharing.
In practice, this means training teams to think before they type. A prompt that includes confidential client names, internal project codes, or unredacted financial data does not just risk a poor answer – it risks a data incident. Good habits include defining the task explicitly, setting constraints on what the AI should not assume, anonymising sensitive values before submitting, and documenting prompts that work so teams can build on them consistently.
Validate Before You Trust
AI models hallucinate. They generate plausible-sounding content that is factually wrong. In low-stakes tasks, this is inconvenient. In regulated industries, it is dangerous.
The EU AI Act’s obligations for high-risk AI systems came into full effect on 2 August 2026, imposing requirements around accuracy, robustness, human oversight, and transparency – raising the compliance bar for organizations deploying AI in high-risk contexts. That regulatory reality should sharpen every organisation’s approach to output validation.
The response is not to distrust AI entirely – it is to build validation into your workflows as a standard step, not an afterthought. Before any AI-generated content is acted upon, shared externally, or fed into another system, a human checkpoint should apply. A simple three-question test helps: Does this align with what I already know? Can I verify the specific figures cited? Would I be comfortable putting my name on this as-is?
Draw Clear Data Boundaries
One of the most pressing responsible AI challenges in 2026 is not about model behavior – it is about what employees feed into models in the first place. The Zscaler data makes this vivid: writing assistants and general-purpose chat tools were among the top drivers of sensitive enterprise data transfers to AI tools, with employees routinely including information without considering what they were sharing.
Organizations need data classification policies that extend explicitly to AI tool usage – defining what data is permitted in AI prompts, which tools are approved for which data classifications, and how to handle tasks that cannot be completed safely. For organizations using Microsoft 365 Copilot or similar enterprise AI platforms, this also means auditing permissions to ensure AI access to organisational data reflects your actual governance posture, not just default settings.
Embed Ethical Judgment Into Daily Practice
Policy documents do not build responsible AI habits. Daily practice does. The most effective organizations are not those with the longest acceptable use policy – they are the ones where ethical judgment about AI has become second nature.
Research across the industry consistently finds that a significant proportion of organizations have yet to put formal guardrails in place to guide AI deployment and operation. That leaves a significant number relying entirely on individual discretion – an unreliable safety net at scale. Ongoing team conversations, not one-time training modules, are what close that gap.
Leaders set the tone. When a CIO or CTO visibly applies critical thinking to AI outputs and acknowledges the limits of these tools, it signals to the entire organization that caution is a professional strength, not a weakness.
Governance Is a Competitive Advantage
Responsible AI is not a brake on productivity. It is the foundation that makes productivity sustainable. Structured approaches like Insentra’s AI Momentum Framework give organizations a clear path to building these habits at scale – moving from ad hoc AI use to governed, repeatable practice. Organizations that build strong AI habits now will be the ones that scale confidently – without the incidents, regulatory exposure, and reputational damage that come from moving fast without guardrails.
The window to shape your organisation’s AI culture is open right now – before behaviors calcify, before incidents occur, and before regulators come knocking. Move thoughtfully, and you will move further.
Continue the Series: A Leader’s Guide to Building an AI-Ready Organisation
Responsible AI habits are one part of a much bigger challenge: building an organization where people have the skills, confidence and guardrails to use AI effectively.
This article is part of our A Leader’s Guide to Building an AI-Ready Organisation series. Explore the other articles:
- You Bought the AI. Now What? Why People Readiness Is the Real ROI Problem — why AI ROI depends as much on people readiness as it does on the technology.
- A Practical Framework for Assessing AI Literacy Across Your Workforce — how to understand where AI capability exists today and where the most important gaps are.
- Building Organisation-Wide AI Literacy From the Ground Up — practical steps for developing AI literacy across roles, teams and levels of your organization.
- How Structured AI Experimentation Turns Passive Awareness Into Daily Practice — how structured, low-risk opportunities to experiment can help employees move from understanding AI to applying it in their everyday work.
- Why AI Skill-Building Must Be Built Around Roles, Not Rolled Out Across Them — why effective AI capability building needs to reflect the tasks, workflows and decisions of different roles rather than relying on one-size-fits-all training.
Together, the series provides leaders with a practical path from AI readiness and literacy to experimentation, role-based capability and responsible everyday use.
Ready to Turn AI Potential Into Organisational Momentum?
Buying AI is easy. Building the capability to use it confidently, responsibly and at scale is where the real work begins.
Insentra’s AI Momentum Framework helps you move from scattered experimentation to purposeful, governed AI adoption aligning your people, skills, use cases and guardrails so AI becomes part of how work gets done, not just another tool people have access to.
Whether you need to understand your current AI readiness, identify capability gaps, strengthen governance or create a practical roadmap for adoption, AI Momentum gives you a structured path forward.
Don’t just deploy AI. Build an organization ready to get value from it.
Explore AI Momentum and start building your AI-ready organisation
