Almost half of office workers are utilizing artificial intelligence tools not provided by their employers, with nearly a third keeping their usage a secret, according to findings from Ivanti’s 2025 Technology at Work Report. The report reveals that 36% of employees believe these tools give them a competitive edge, while 30% express concerns about job security. Notably, 42% of employees reported using AI in their daily tasks, a significant rise from just 25% the previous year. Ivanti cautions that this covert use of AI tools poses serious risks to organizations, including potential breaches of company policies and data privacy regulations.
A recent article from The New Stack emphasizes that artificial intelligence will not resolve data modeling issues, highlighting the crucial need for organizations to prioritize data quality at the operational level. As generative AI applications require real-time data, the traditional approach of cleaning data after it has been collected is no longer viable; incorrect or poorly structured data leads to faster but misguided decisions. The article notes that the responsibility for data modeling is shifting from analytics teams to operational layers, where real-time decisions are made. According to Microsoft Research, using structured knowledge graphs can significantly enhance the performance of generative AI systems, improving response accuracy from 16% to over 72%. This trend underscores the importance of having robust data models to ensure that AI-driven decisions are both reliable and aligned with business logic.
A recent study by Giskard, an AI testing company based in Paris, reveals that asking AI chatbots for concise answers can increase their tendency to hallucinate. The researchers found that prompts requesting shorter responses, especially for ambiguous questions, negatively impact the factual accuracy of AI models. The study highlights that models like OpenAI’s GPT-4o and Anthropic’s Claude 3.7 Sonnet exhibit reduced accuracy when instructed to be brief. The researchers noted that when models are limited in their responses, they often prioritize brevity over factual correctness, making it challenging to address misinformation effectively. Giskard’s findings suggest that seemingly innocuous instructions, such as “be concise,” can undermine an AI’s ability to clarify false premises in user queries.
Oh, and those discussions about energy use of AI? A recent analysis reveals that using ChatGPT has a minimal carbon footprint, challenging the notion that artificial intelligence tools significantly harm the environment. For instance, data suggests that a single query to ChatGPT uses approximately three watt-hours of electricity, which represents only 0.00007% of the annual electricity consumption per person in the United Kingdom. Moreover, the carbon emissions associated with a year of daily ChatGPT usage amount to just 11 kilograms of carbon dioxide, a negligible increase compared to the average individual’s carbon footprint of seven tonnes. The analysis, conducted by Andy Masley, indicates that even if ChatGPT is ten times more energy-intensive than a Google search, the overall environmental impact remains small, especially when compared to other lifestyle choices that contribute significantly to carbon emissions. This data underscores the argument that individual use of ChatGPT is not detrimental to the environment, suggesting that users can engage with AI tools without guilt.
Why do we care?
AI is outpacing governance, data readiness, and user expectations. This is also a roadmap for where AI support needs to mature fast. Policy and ethical use training. Governance over shadow AI use. Data structure validation and modeling. Responsible AI UX design. Long-term sustainability planning.
Many organizations still lack any AI policy. If that continues, shadow usage will outpace governance—leading to breaches, hallucinated outputs, and trust breakdowns.
This is a moment for IT service providers to step up—not just as tool implementers, but as AI stewards who understand risk, structure, and human behavior. That’s the path to relevance and resilience.