Generative AI tools entered the workplace faster than most companies could write policy for them. Employees began pasting client data, source code, and internal documents into ChatGPT and similar tools within months of public release, often without any review process in place. Regulators have taken notice: the FTC, HIPAA enforcement bodies, and state privacy regulators in California have all signaled that AI use does not exempt a company from existing data protection obligations. For SMBs in healthcare, SaaS, technology, and defense contracting, AI security compliance is no longer optional groundwork. It is a baseline expectation from clients, auditors, and regulators alike.
Below are the questions Planet 9 hears most often from founders, CTOs, and IT managers working through this problem.
How do I create an AI acceptable use policy for employees?
An AI acceptable use policy is one of the foundational documents in any AI governance program. It should state plainly what employees can and cannot do with AI tools, specific enough to leave no ambiguity about acceptable behavior.
The policy should start by identifying which AI tools are approved for processing company data. Personal accounts, free trial accounts, or unsanctioned tools should never handle company-specific content, particularly anything involving sensitive data. Employees will experiment with new tools regardless of policy, and that exploration has value. The line that matters is keeping company sensitive data away from unapproved accounts.
Even approved tools require clarity around data classification. Some organizations process PII or protected health information, and certain tools may be cleared for that sensitivity level while others are not. Employees need to know exactly which tools are authorized for which data types. Login credentials, passwords, API keys, and similar secrets should never be entered into any AI tool.
Where the platform allows it, model training and data retention settings should be disabled. Employees also need training on output review: AI-generated content can contain errors, and a human should validate any output before it is used. Code generated by AI tools carries the same requirement, reviewed and tested for security vulnerabilities before deployment rather than assumed to be production-ready.
Finally, the policy should state that AI tools cannot be used to generate security exploits, bypass security controls, produce synthetic media such as deepfake audio or video, or reproduce copyrighted material. Once drafted, the policy needs sign-off from management, and every employee should review and formally acknowledge their commitment to follow it. This acknowledgment step matters for regulatory compliance AI documentation and gives the organization a clear audit trail.
What are the compliance risks of using ChatGPT and generative AI at work?
The risks here are substantial and often underestimated. When a company handles sensitive data such as PII or PHI, using a third-party tool requires a contractual agreement ensuring the provider secures that data appropriately. Using ChatGPT or comparable tools without such an agreement can put a company in direct violation of HIPAA, GDPR, or other data privacy and security regulations governing its industry.
Because these tools are so easy to access, employees who have not received proper training can input sensitive information without realizing the regulatory implications. A staff member drafting a patient communication or summarizing a contract more efficiently may not connect that action to a compliance violation. The consequences can include data breaches, regulatory penalties, loss of client contracts, and lasting reputational damage.
This is the core argument for structured employee education. Once staff understand which tools are approved and what data is permitted with each one, the likelihood of accidental violations drops considerably.
How do you prevent sensitive data from leaking into ChatGPT or other AI tools?
Preventing data leakage into AI tools generally requires two parallel approaches: technical controls and awareness training.
On the technical side, Data Loss Prevention (DLP) tools, properly configured on employee workstations, can detect attempts to upload or paste sensitive information into AI platforms or unauthorized SaaS applications, and block the action before it completes. This creates a real-time backstop that does not depend on an employee remembering policy in the moment.
On the human side, most policy violations happen without malicious intent. Employees who have not been trained simply do not recognize the boundary they are crossing. Structured education on proper AI tool use, paired with clear documentation of which tools handle which data types, reduces violations because people understand the reasoning behind the rule rather than just the rule itself.
In short: DLP tools detect and prevent unauthorized data input at the technical layer, while security awareness training builds the judgment needed to avoid the situation in the first place. Together, these form the practical foundation of data privacy and security in AI programs for growing organizations.
Does your company need ISO 42001 or NIST AI RMF compliance?
Neither framework is currently mandated by law in the United States, and no jurisdiction requires a company to hold ISO 42001 certification or formally adopt the NIST AI Risk Management Framework. That said, the regulatory picture is shifting in ways that make voluntary adoption more than a nice-to-have. Colorado's original AI Act once offered companies a safe-harbor defense for following NIST AI RMF or ISO 42001, but that law was repealed and replaced before taking effect, and the safe harbor did not carry over. Organizations that assumed framework adoption would shield them from state-level AI liability need to revisit that assumption. The EU AI Act, meanwhile, requires risk-based governance for high-risk AI systems starting in August 2026, and companies doing business in the EU face mandatory obligations that closely resemble what ISO 42001 and NIST AI RMF already cover.
Even without a direct legal mandate, adopting one of these frameworks gives an organization a structured way to manage AI security and compliance risks rather than addressing issues reactively as they surface. ISO 42001 is also becoming a procurement requirement in practice, with enterprise customers and federal contracting officers increasingly asking for it as a condition of doing business, even where no statute requires it.
These frameworks provide the foundation for a comprehensive AI governance program, accounting for the range of risks AI technology introduces rather than treating AI as a single, undifferentiated tool. For SMBs pursuing SOC 2 readiness or preparing for due diligence with enterprise clients, voluntary adoption of a recognized framework often signals maturity that clients and auditors look for, and it positions the organization ahead of regulatory requirements that continue to expand.
How do companies audit and govern AI systems for security risks?
This is where frameworks like ISO 42001 or NIST AI RMF provide practical value. They help organizations structure their approach to AI governance and compliance risk across three distinct areas.
The first is general employee AI use: establishing an acceptable use policy, training staff, and defining acceptable and unacceptable use within the organization. The second is third-party AI service providers: putting proper agreements in place, conducting due diligence on AI model vendors, and monitoring their ongoing compliance to confirm contractual security obligations continue to be met. The third is internal AI development. This may not apply to every company today, but a growing number of technology companies will eventually build their own AI models or fine-tune existing ones. Policies covering internal development need to address security and compliance risks alongside AI-specific concerns such as model drift, ethical considerations in decision-making, and the handling of special categories of data.
Treating these three areas as separate workstreams, rather than folding them into a single generic policy, tends to produce a more durable AI governance risk and compliance program because each area carries distinct obligations and stakeholders.
Who is responsible for AI security in an organization?
The Chief Information Security Officer, or an equivalent role such as Head of Security or VP of Security, typically owns AI security within an organization. AI risk extends beyond the security function, however. Significant legal exposure exists as well, particularly for organizations developing their own AI models or processing regulated data types.
For this reason, AI governance should not sit exclusively within the security organization. A more effective structure brings together stakeholders from IT, security, legal, and HR through an AI compliance committee, addressing the full range of risks AI introduces rather than relying on a single department to anticipate every angle. This cross-functional approach tends to catch gaps that a security-only review would miss, particularly around contractual obligations and employment-related AI use cases such as hiring tools or performance monitoring.
AI security compliance is becoming a standard expectation for SMBs in healthcare, SaaS, technology, and defense contracting, not a future consideration. Organizations that build acceptable use policies, train employees, and establish clear governance structures now will be better positioned for client due diligence, audit readiness, and regulatory scrutiny as AI oversight continues to develop.
Planet 9 is a Bay Area cybersecurity consulting firm specializing in audit readiness for SMBs in healthcare, SaaS, and technology. Our vCISOs and compliance managers help organizations choose the right approach, configure GRC tools if needed, and get audit-ready without wasted time.
