AI tools are moving faster than the policies designed to govern them. It’s important for businesses in San Francisco to understand the controls around how AI tools are used. In this blog, we cover the key security and compliance considerations every organization should address before and after introducing AI into daily operations.

AI Risk Management Starts Before You Deploy

AI systems have defining characteristics that traditional business software typically does not. They process, and in some cases externally transmit, the data fed into them. When that data includes client records, financial information, or proprietary business processes, the implications for confidentiality and compliance are significant. California’s data privacy obligations under the CCPA add another layer to that exposure, and regulated industries carry additional requirements beyond that.

According to IBM’s 2025 Cost of a Data Breach Report, organizations that allowed high levels of shadow AI – employees using unauthorized AI tools without IT oversight – incurred an average of $670,000 more per breach than those that kept shadow AI to a minimum. The same report found that 63% of breached organizations either had no AI governance policy or were still in the process of developing one.

This is important if your business is in legal, financial services, or insurance, as those figures carry direct operational weight. Introducing AI tools without first reviewing what data those tools can access and under what conditions creates compliance exposure that is difficult to reverse.

The Shadow AI Problem and Why AI Security Policies Matter

Shadow AI describes the AI tools employees adopt without the knowledge or approval of the IT team. It is a more widespread problem than most organizations realize. According to LayerX Security’s research, 45% of enterprise employees are already using generative AI tools, and 67% of that usage happens through unmanaged personal accounts, outside any enterprise oversight. Of the files employees upload directly into AI platforms, 40% contain PII or payment card data.

Data entered into consumer-facing AI tools may be retained for model training, stored outside the organization’s control, or exposed in a future breach of the AI provider. Once it’s out, it cannot be recalled.

A Palo Alto Networks analysis of generative AI usage across 7,000 enterprise customers found that the average organization has around 66 generative AI applications in active use, with 10% classified as high risk. Data loss prevention incidents related to generative AI more than doubled in early 2025.

An AI security policy closes this gap by specifying which tools are approved, who can use them, what data categories are off-limits, and how violations are handled. Without one, those decisions get made informally, which means they often don’t get made at all.

AI Governance for Businesses Goes Beyond a Policy Document

AI governance means putting the operational structure in place to enforce that policy consistently.

That includes identity and access management, ensuring employees can only reach the AI tools and data their role requires. Audit trails capture how AI tools are being used, though IBM’s 2025 data suggests that most organizations skip this step. Of those with governance policies already in place, only 34% were conducting regular audits to detect unsanctioned AI use. A clear vetting process ensures new tools are assessed before adoption, rather than flagged as a problem after the fact.

The NIST AI Risk Management Framework provides a structured approach organizations can use to assess and manage AI-related risk across four functions: Govern, Map, Measure, and Manage. It’s designed to be adaptable regardless of sector or organization size, and NIST has published a dedicated profile for generative AI.

The Cybersecurity Infrastructure Behind Secure AI Implementation

Governance works only as well as the technical controls behind it. Secure AI implementation depends on the right cybersecurity infrastructure as the foundation that makes policy enforceable.

Endpoint protection limits what software can be installed or accessed on company devices, which is the most direct defense against employees connecting personal AI accounts to work systems. For a law firm whose case files sit on shared drives, or an insurance company whose staff handle claims data daily, that control makes a meaningful difference. Identity and access management restricts which systems and data each user can reach, reducing the exposure window if an AI tool is compromised. Network monitoring and threat detection provides visibility into data flows, making it possible to spot unusual patterns, such as large volumes of data moving toward an external AI platform, before they escalate.

This is the managed IT infrastructure that should already be in place, and that becomes even more critical once AI is part of the picture.

How Centarus Supports Responsible AI Adoption for San Francisco Businesses

Centarus works with San Francisco and Bay Area businesses across legal, financial services, and insurance, where the stakes around data handling are particularly high and where regulators have specific expectations about how client information is managed.

Our approach to AI adoption follows a structured process. We start with a full assessment of how your business operates, moving through AI implementation with real governance and security controls built in from the start. We continue with ongoing support that keeps your systems monitored, secure, and aligned with your business as it evolves. That structure means AI gets addressed as part of a broader security and technology strategy, rather than a standalone project that sidesteps the compliance and data-handling requirements your industry carries.

For businesses beginning to think seriously about AI, the right question isn’t whether to adopt it. It’s whether the security infrastructure is in place to do it without creating exposure you can’t afford.

If you’re unsure where to start, book a complete consultative discovery conversation with Centarus. We’ll help you understand exactly what your current IT setup can and can’t support and what it would take to close the gaps.

CTA

Headline: Is Your AI Adoption Leaving the Door Open?
Subline: Book your complete consultative discovery conversation today.

FAQs

What types of data should employees never put into AI tools?
Anything your business would be uncomfortable with seeing in a data breach: client names, financial records, case notes, contract terms, employee information, and anything subject to HIPAA, CCPA, or attorney-client privilege. Consumer AI tools often retain input data and may use it for model training, and once it’s shared in a prompt, it can leave your organization’s control permanently.

How does California’s CCPA affect AI use in my business?
The CCPA gives California consumers the right to know what personal data a business holds about them, request its deletion, and opt out of its sale. If employees are feeding customer data into AI tools that store or process it externally, that creates potential exposure around data retention and deletion rights. Any AI vendor handling personal data should have a data processing agreement in place.

What’s the difference between an AI security policy and general IT security?
General IT security protects your systems from external threats. An AI security policy governs how people inside your organization use AI tools: which ones are permitted, what data can be shared, and who approves new tools. A business can have strong perimeter security and still have employees pasting client records into consumer AI tools without anyone knowing.

Do small businesses really need formal AI governance?
A one-page list of approved tools, a clear rule about what data is off-limits, and a simple process for flagging new tools before they’re adopted is enough to start. The cost of a compliance violation or data incident is disproportionately high for smaller organizations.