Billions of daily queries, zero control—your data is training tomorrow’s competitors
In 2025, enterprise users—particularly knowledge workers in sectors like IT, finance, and professional services—are leveraging public AI engines (such as ChatGPT, Claude, or Gemini) for information retrieval at a remarkably high frequency, often through natural language queries for tasks like research, summarization, and quick analysis, despite the inherent risks tied to these models’ data practices.
Frequency of Use
Adoption has surged, driven by the accessibility and productivity gains of these tools, even as many organizations lag in deploying sanctioned enterprise versions. Key statistics from recent surveys paint a picture of routine, high-volume engagement:
- Daily Usage: Approximately 37-52% of professionals, especially higher-earning ones (over $125,000 annually), interact with public LLMs daily for work-related queries. This includes shadow AI practices, where employees bypass IT oversight; in over 90% of companies, workers report using personal accounts multiple times a day, every workday, for information finding—far outpacing official pilots.
- Weekly Usage: An additional 46% engage several times per week, with only 16.7% using them less than once weekly, indicating that for most adopters, it’s a habitual tool woven into daily workflows. Overall, 68% of enterprise employees access public AI via personal accounts, and 40% of U.S. workers report any AI use at work, doubled from 2023.
- Context for Information Finding: These interactions often focus on generative queries (e.g., “Summarize market trends for Q3” or “Analyze this dataset for insights”), with 70% preferring public tools for quick research due to their familiarity and iterative capabilities; usage has shifted toward automation-dominant tasks, comprising 50% of sessions by mid-2025. Globally, this equates to billions of enterprise-adjacent queries annually, as 78% of organizations now incorporate AI in at least one function.
This “shadow economy” thrives because public engines deliver immediate value—saving hours on routine info retrieval—while enterprise rollouts remain mired in integration challenges, leading to 90%+ worker adoption rates in surveyed firms.
Related Risks
The core vulnerability stems from how public AI engines like those from OpenAI or Anthropic process and retain user inputs: many still use query data to refine and train models (unless explicitly opted out via enterprise agreements), potentially exposing proprietary information to unintended reuse. This creates a cascade of enterprise-specific hazards, amplified by the sheer volume of daily queries:
Risk Category | Description | Prevalence/Impact in 2025 |
Data Privacy & Leakage | Sensitive inputs (e.g., client details, strategies) entered into public tools can be logged, retained (up to 5 years in some cases), or ingested into training datasets, violating GDPR, HIPAA, or CCPA; 57% of employees admit feeding confidential data into these systems, creating governance blind spots. | AI incidents rose 56% in 2024 alone, with privacy breaches topping concerns; public data may resurface in outputs for others, eroding trust. |
Intellectual Property (IP) Theft | Unique queries or proprietary insights could train competitors’ models indirectly, as terms often grant broad usage rights for “service improvement”; no direct profit for the enterprise, but potential competitive dilution. | Cited by 38% of S&P 500 firms as a top AI risk, with 7 in 10 large U.S. companies disclosing related exposures. |
Compliance & Legal Exposure | Retained prompt histories may be subpoenaed in discovery, or inaccurate outputs could lead to faulty decisions; 44% of IT leaders flag this as the biggest adoption barrier. | Regulatory scrutiny intensifies, with reputational damage affecting 38% of disclosures; fines could reach millions under data laws. |
Cybersecurity & Reputational Harm | Enlarged attack surfaces via API integrations or phishing via AI-generated content; 20% of firms report cyber risks tied to public tools. | 62% of users wrongly assume interactions aren’t stored, heightening breach potential; shadow use evades audits, compounding issues. |
Mitigation strategies include zero-trust enterprise LLMs, data anonymization, and policy enforcement, but with daily habits entrenched, full containment remains elusive—underscoring the tension between AI’s speed and security’s demands.
Celiveo 365 AI-powered Documents Management has been designed with Zero-Trust-Access and High Security at its core, discover how it protects your company and prevents data leaks at https://celiveo.com
