What Your Employees Don't Know About AI Tools Could Cost You
Your employees are using AI tools to get their work done faster. That's mostly a good thing. But the data they're feeding into those tools? That's where it gets complicated.
The Problem Nobody's Talking About Loudly Enough
Right now, somewhere in your organization, an employee is pasting client data into ChatGPT to write a summary. Another is uploading a financial spreadsheet to an AI tool to generate analysis. A third is using a free AI tier, one that explicitly trains on user inputs, to draft a legal communication.
None of them are doing it maliciously. They're doing it because it works, nobody told them not to, and there's no policy saying they shouldn't.
The gap isn't behavior, it's documentation. Most employees would follow a clear policy. The problem is the policy doesn't exist yet.
Real Scenarios That Happen Every Week
Scenario 1, The Helpful Summarizer
A sales rep pastes a full client contract into ChatGPT to get a quick summary before a call. The contract contains PII, confidential terms, and proprietary pricing. That data is now in a third-party AI system, potentially stored, potentially used for model training.
Scenario 2, The Efficiency Win Gone Wrong
An HR manager uploads a spreadsheet of employee salary data to an AI tool to "clean it up and analyze patterns." The tool is a free tier product. The terms of service allow the provider to use inputs for training. Employee compensation data is now training data.
Scenario 3, The Developer Shortcut
A developer pastes proprietary source code into an AI coding assistant to debug a function. The code contains business logic that represents years of competitive advantage. It's now sitting in a third-party system with unclear retention policies.
Why Free AI Tiers Are a Specific Risk
Many AI tools operate on a tiered model: the free version is funded partly by using your inputs as training data. This isn't hidden, it's usually buried in the terms of service that nobody reads.
OpenAI, Google, and most major AI providers offer opt-out options for enterprise plans but not free tiers. If your employees are using personal ChatGPT Free accounts for work, your company data may be improving a competitor's AI model.
The fix is simple: require company-licensed accounts for AI tool use, and specify this in your AI Acceptable Use Policy.
What's at Stake
The consequences of unmanaged AI tool use range from embarrassing to catastrophic:
Data breaches, PII, PHI, or financial data exposed through third-party AI systems
Regulatory penalties, GDPR violations can cost up to 4% of global revenue; HIPAA fines reach $1.9M per violation category annually
Client trust damage, A client discovering their data was fed to an AI tool without consent is a relationship-ending conversation
Intellectual property exposure, Proprietary code, formulas, and strategies shared with AI tools may not stay proprietary
AI-generated misinformation, Employees sending AI-generated content to clients without review creates liability for inaccurate or fabricated information
The Three Things You Need to Do Right Now
1. Create an AI Acceptable Use Policy
This doesn't need to be 50 pages. A clear, plain-English policy that defines approved tools, prohibited data inputs, human review requirements, and consequences for violations is what you need. Employees can't follow a policy that doesn't exist.
2. Approve Specific Tools (and Disallow Everything Else)
Build a list of approved AI tools, their approved use cases, and whether employees may use personal accounts or only company-licensed accounts. "Use AI tools responsibly" is not a policy, named tools with defined permissions is a policy.
3. Train Your Team
Policy without training is a document nobody reads. A short 15-minute session covering what can and can't go into AI tools, what the approved tools are, and what happens if someone violates the policy goes a long way. Repeat it annually.
Your AI policy, done in under an hour.
Our AI Acceptable Use Policy template covers approved tools, data classification rules, prohibited uses, human review requirements, disclosure obligations, and enforcement, written in plain English with fill-in-the-blank placeholders throughout.
AI tools aren't going away, and you shouldn't want them to. The productivity gains are real. But "we didn't have a policy" is not a defense you want to use with a regulator, a client, or a judge.
The window to get ahead of this is now, while AI governance is still being established rather than enforced. Organizations that put clear policies in place today will be in a far stronger position than those scrambling to respond after an incident.
Your employees don't know what they don't know. That's your job to fix.
📬
Get CMMC tips and template updates
No spam. Just practical guidance on CMMC compliance and new resources when we publish them.
Frequently Asked Questions
CISA's 'Guidance for Organizations Using AI Tools' (2024) identifies four primary risk categories: data exposure (sensitive data entered into AI systems may be retained or disclosed), prompt injection attacks (malicious inputs that manipulate AI behavior), supply chain risks (vulnerabilities in underlying models or APIs), and over-reliance on AI outputs without human verification. NIST AI 100-1 additionally identifies bias, privacy, and accountability risks specific to organizational AI deployment.
It depends on the tool and terms of service. Many consumer-grade AI tools, particularly free tiers, historically used conversation data for model improvement unless users opted out. Enterprise tiers typically offer stronger data use restrictions and data processing agreements. CISA's 2024 guidance recommends that organizations establish an approved AI tool list with data handling terms verified and documented before any tool is deployed for work use.
NIST SP 800-218A (Secure Software Development Practices for Generative AI) and NIST AI 100-1 both emphasize the need for organizational AI governance including documented acceptable use policies, an inventory of AI systems in use, and formal risk assessment before deployment. NIST also recommends treating AI tools as third-party software subject to supply chain risk management practices under NIST SP 800-161, given dependencies on external models and APIs.
Prompt injection is an attack where malicious instructions embedded in content processed by an AI system cause it to behave unintentionally, for example, instructions hidden in a document that cause an AI assistant to exfiltrate data or take unauthorized actions. NIST AI 100-2 (Adversarial Machine Learning) and the OWASP LLM Top 10 both list prompt injection as a critical AI security risk. Mitigations include input validation, privilege separation, and human review of AI-generated actions before execution.
Yes. CISA's AI security guidance and NIST AI 100-1 both recommend security awareness training as a core control for AI risk management. Employees should understand what data is appropriate to enter into AI tools, which tools are organizationally approved, how to identify AI-generated misinformation or deepfakes, and how to report AI-related security incidents. This training should be documented as part of the organization's formal security awareness program.