Shadow AI Rife as Half of Workers Hide Use From Management
A recent study indicates a significant prevalence of employees secretly using artificial intelligence within their workplaces, a practice termed "shadow AI". Research found over half of all workers admit to concealing their AI-assisted tasks from management. This trend raises concerns regarding data security, compliance, and overall corporate oversight.
According to an article in *Forbes* by Lance Eliot, a global study conducted by KPMG and the University of Melbourne discovered that 57 per cent of more than 48,000 respondents acknowledged hiding their AI usage at work. This clandestine activity is largely driven by a lack of trust in the organisation, alongside fears of job replacement, potential stigma, or the perception of being less capable. Firms often implement stricter governance, such as usage policies and monitoring, which may inadvertently encourage staff to further obscure their methods.
But the issue extends beyond mere defiance. When employees eventually reveal successful AI workflows, they may face punitive measures rather than commendation. This discourages open dialogue and prevents organisations from identifying potentially beneficial innovations. The problem of hidden AI usage presents considerable downsides for companies, which may remain unaware of critical risks.
Organisations typically categorise AI use in two ways: aboveboard or under-the-table. Aboveboard use occurs openly, with the firm aware of the AI deployment. Some companies instruct workers to use AI, while others allow staff to volunteer its use. In these instances, firms may reward innovative AI applications or simply permit them without formal endorsement. However, workers introducing AI independently bear the risk of adverse outcomes, potentially affecting promotions or employment if their methods fail.
The use of AI can produce both favourable and undesirable results. Even officially sanctioned AI can fail, with workers sometimes relying too heavily on its output, neglecting to override or correct errors. A notable example involved "token-maxxing," where some companies measured AI usage by token or word consumption. Some employees exploited this metric, incurring substantial costs for the firm by having AI perform redundant tasks, merely to increase token counts. So, if AI use goes awry in plain sight, a company has a chance to detect the problem and act, either stopping or adjusting its application.
However, detecting under-the-table AI use is significantly more difficult. If hidden shadow AI negatively affects a company, pinpointing AI as the cause can be a substantial challenge. An employee determined to conceal their AI use can maintain secrecy effectively. Leadership might then incorrectly attribute blame, focusing on visible culprits rather than identifying AI as a potential root issue.
Reasons for workers to conceal AI usage vary. Employees may worry that colleagues or supervisors will perceive them as lacking ability or needing AI to perform their duties. This perception could hinder career progression within the firm, leading workers to keep their AI assistance private. Another factor is the negative term "AI slop," which refers to generative AI producing low quality content. Staff fear that using AI for reports, for example, could lead to their work being labelled sloppy, regardless of the actual output quality.
One primary fear among workers involves job security. Demonstrating that AI can perform a substantial portion of their tasks might lead to the company replacing staff with AI solutions. An innovative employee could inadvertently jeopardise their own position by proving AI's capabilities. This can be a harsh consequence for those seeking to introduce new methods. And the potential for legal complications arises if an employee takes proprietary AI solutions or company data used for AI training to a new employer.
Organisations must adopt a balanced approach. Penalising shadow AI merely pushes such practices further underground. A company needs to foster an environment where trust is paramount, avoiding punitive reactions that could escalate into dismissals. This calls for a "Goldilocks principle" for shadow AI. Leadership should not encourage unchecked, under-the-table AI, which might introduce viruses or privacy breaches.
But the process for approving aboveboard AI usage must also not be overly complicated. Onerous bureaucratic procedures can deter employees, forcing them towards unapproved AI solutions. Streamlining approval pathways for AI implementation is crucial to prevent workers from circumventing official channels. Executives who claim no shadow AI exists in their organisations may be mistaken. Investigations often reveal numerous instances of covert AI use, even in firms with strict policies.
A significant percentage of employees use artificial intelligence covertly at work.
Lack of trust and fears of job displacement contribute to this concealed usage.
Organisations risk undetected problems and loss of intellectual property due to shadow AI.
Companies should balance formal governance with accessible approval processes for AI tools.
Source: Forbes
