The Reality of “Shadow AI” Spreading Across the Workplace
“Two out of three IT managers report an increase in shadow AI”—this finding from an ASCII.jp survey highlights a management challenge that can no longer be ignored. The “shadow AI” phenomenon, where employees start using ChatGPT and other generative AI tools without official approval, underscores the dilemma between security risks and operational efficiency.
At the same time, news from Yahoo! Japan reporting that “even major AI companies are cutting budgets and imposing usage limits” reveals a reality where the operational costs of generative AI are beginning to outpace revenue. If large corporations are struggling with these costs, how will small and medium-sized enterprises be affected? This article explores the true sentiments on the ground and the countermeasures leaders should take.
Three Factors Behind the Rise of Shadow AI
1. A Genuine Need for Greater Efficiency
Employees on the front lines constantly feel the need to work more efficiently. Generative AI excels in areas like drafting emails, summarizing documents, and analyzing data—tasks that are part of daily work. However, the slower a company is to officially adopt these tools, the more employees will start using them with personal accounts.
At one of my consulting clients, a sales manager told me with a wry smile, “My subordinate was using ChatGPT without permission to create proposals.” While productivity did improve, the reality is that we can’t simply celebrate this, given the risk of customer data being sent to external servers.
2. The Perception Gap Between IT and the Front Line
According to the ASCII.jp survey, about 67% of IT department managers feel the increase in shadow AI. On the other hand, awareness among top management tends to be lower. Many executives still view “AI adoption as a future issue.”
This perception gap accelerates unauthorized use on the ground. Without company-wide AI guidelines, employees end up choosing tools and deciding how to use them based on their own judgment.
3. Balancing Cost Burden and Profitability
As reported by Yahoo! Japan, even major AI companies are starting to see operational costs exceed revenue. API usage fees for ChatGPT can vary, and it’s not uncommon for monthly bills to jump from a few hundred to several thousand dollars.
The CTO of a mid-sized logistics company lamented, “When we rolled out ChatGPT company-wide, our monthly bill exceeded $2,000.” While the results were tangible, the cost-benefit analysis had been insufficient.
Three Risks Posed by Shadow AI
1. Risk of Data Leakage
The most serious risk is the potential for customer information and confidential data to be sent to external AI servers. With the free version of ChatGPT, input data may be used for training, making it a major problem to enter contracts or personal information.
When I automated AI-based contract review, I always designed the system to use a paid API that does not use data for training. Overlooking this point can lead to significant trouble down the line.
2. Compliance Violations
In some industries, the use of external AI services itself may be subject to regulation. For financial institutions and healthcare providers, sending customer data externally could violate the law.
The more shadow AI proliferates, the higher the risk of such compliance violations. AI usage that IT departments cannot track can become a blind spot in audits.
3. Promoting Over-Reliance on Specific Individuals
Shadow AI can lead to a situation where only certain employees possess advanced AI skills, creating a dependency. If that employee leaves, the know-how is lost, and operations can stall.
In one IT company, after a young employee skilled in AI left, no one could manage the automation scripts they had built, leading to a system outage. These are the hidden risks behind shadow AI.
Three Concrete Steps for Management
1. Establish Company-Wide AI Guidelines
First, clarify the company’s rules for AI use. Specifically, guidelines should include the following items:
- A list of approved AI tools
- Types of data that can be input (prohibiting personal and confidential information)
- An application process for use
- Rules for information management
In my experience, simply communicating these guidelines to all employees can make about 70% of shadow AI visible. Unauthorized use happens precisely because there are no rules.
2. Introduce AI Tools Through an Approval System
Instead of a “ban,” using an “approval system” allows you to address on-the-ground needs while ensuring security. For example, the following workflow is effective:
- Employees apply for the AI tools they want to use
- IT conducts a security assessment
- Management evaluates the cost-benefit
- Only approved tools become available
In one manufacturing company, implementing this workflow reduced shadow AI by 80% within six months. At the same time, it made visible the tools that the front line truly needed, speeding up adoption decisions.
3. Start Small with AI Adoption and Manage Costs
Rather than a company-wide rollout, it’s more realistic to start small, on a department-by-department basis. In terms of cost, here are some options:
- ChatGPT Plus ($20/month per person): Ideal for individual use
- Claude Pro ($20/month per person): Strong for long-form text processing
- In-house API setup (starting from a few hundred dollars/month): For large-scale use
In my own company, I use three AIs (Claude, ChatGPT, and Grok) in combination, spending about $140 per month to generate value equivalent to roughly $50,000 per year. Starting small, verifying results, and then scaling up is the way to avoid failure.
Conclusion: Turning Shadow AI into a Management Asset
The rise of shadow AI is a reflection of the “true feelings” on the ground: employees recognize the power of AI and want to use it to improve their work. Instead of viewing this solely as a risk, it’s crucial to see it as a management resource to be leveraged.
The ASCII.jp survey’s finding of a “2.7x gap in AI adoption between large and small companies” suggests this disparity may widen further. Rather than ignoring shadow AI, you can safely accelerate on-the-ground innovation by establishing proper guidelines and managing it through an approval system.
On the cost front, instead of a company-wide, one-time rollout, a small-scale introduction tailored to each department’s needs is more practical. Many AI tools are available starting at a few hundred dollars per month. By visualizing ROI and expanding gradually, you can keep unnecessary costs in check.
Rather than fearing the term “shadow AI,” the right approach for companies today is to listen to the true sentiments on the ground and leverage AI as a strategic asset.
Comments