• Regulated companies should organize and protect their data before selecting an AI platform.

• AI agents introduce a different risk profile because they can act on files, systems, emails, workflows, and customer interactions.

• Human review, access controls, backups, logging, and ongoing permission checks should be part of the rollout plan.

• Insurance coverage should be reviewed before AI tools are broadly deployed, because traditional policies may not clearly address every AI related loss.

Many companies are moving quickly to adopt AI. That makes sense. The productivity upside is real, especially for organizations trying to summarize documents, support customers, analyze information, improve internal workflows, or help employees work through large volumes of data.

But for regulated companies, the conversation has to start in a different place.

At Apex Technology Services, we have been consulting with numerous organizations, many of them in regulated markets, about how to deploy AI safely. The pattern is becoming clear. The companies that are approaching AI thoughtfully are not beginning with a model comparison. They are not asking whether Claude, ChatGPT, Gemini, Microsoft Copilot, OpenClaw, or another platform has the most impressive demo.

They are asking a more important question: is our data ready?

That question matters because AI can expose problems that already exist inside an organization. If payroll data is sitting in the wrong folder, AI can make it easier to find. If HR investigation notes are accessible to people who should never see them, AI can surface them faster. If confidential deal documents, legal files, board materials, customer records, financial projections, or market sensitive information are loosely permissioned, AI can turn a messy data environment into a serious compliance issue.

This is why the starting point should be data organization, not AI enthusiasm.

For regulated companies, data needs to be classified, permissioned, and siloed properly. Payroll information should not be available to the broader company. HR investigations should be tightly limited. Confidential deals that could move markets should be separated and protected. Legal matters, customer records, financial data, and regulated communications need the same level of care.

That does not mean locking up everything so employees cannot work. It means creating the right boundaries. The goal is to make sure employees, systems, and AI tools can access what they need, and not what they should not see.

The NIST AI Risk Management Framework is helpful here because it frames AI risk around governance, mapping, measurement, and management. That may sound formal, but the practical meaning is simple. Before AI becomes part of daily operations, companies should understand the use case, the data involved, the people who have access, the risks created, and the controls needed to manage those risks. 

Once the data layer is better organized, the next question is platform selection.

Many clients we speak with have chosen Claude for certain use cases. Others are evaluating ChatGPT, Gemini, Microsoft Copilot, or tools that connect directly into business systems. The right answer may vary by industry, use case, security requirements, and existing infrastructure. A financial services firm may have different needs than a healthcare organization, law firm, manufacturer, government contractor, insurance agency, or accounting firm.

The platform decision should include more than output quality. Companies should review data retention terms, enterprise controls, admin visibility, logging, contractual protections, vendor security posture, model behavior, and integration options. They should also look at how the tool connects to Microsoft 365, Google Workspace, Slack, CRM systems, file storage, ticketing platforms, code repositories, and endpoint environments.

That last point is critical because AI agents change the risk profile.

A chatbot generally answers questions. An agent may take action. It can potentially create, delete, move, overwrite, send, publish, summarize, approve, or update things. In the right environment, that can be useful. In the wrong environment, it can be dangerous.
This is where companies need to be especially careful with tools such as OpenClaw and other agent frameworks. These tools may be powerful, but power is exactly the issue. OpenClaw related security research and industry analysis has highlighted risks involving broad permissions, exposed agents, malicious skills, prompt injection, file access, command execution, and system level compromise. Microsoft’s security analysis warned that indirect prompt injection can occur when agents ingest malicious instructions from external content, while IBM noted that agentic systems can combine browser automation, SSH tooling, file system access, messaging integrations, and an LLM into one expanded attack surface. 

That does not mean companies should avoid agents forever. It means they should not give agents broad authority before controls are in place.

For many regulated organizations, a smarter starting point is an isolated computer, a sandboxed environment, or a read only workflow. Let the team test what the agent does. Watch how it behaves. Review the logs. Confirm what it can access. Make sure backups are running frequently. Then expand carefully.

Human in the loop review should be the default early on, especially for regulated markets.

That is not just a comfort measure. It is a control. Human review should be required before AI sends external emails, changes customer records, modifies contracts, updates financial information, touches production systems, publishes content, creates customer offers, deletes files, overwrites documents, or makes decisions that could affect employees, customers, investors, or regulated obligations.

Even Anthropic’s own Claude Code documentation and engineering materials reflect the importance of permissions. Anthropic states that Claude Code asks users for approval before running commands or modifying files by default, although it also notes that repeated approvals can create approval fatigue. That is an important reminder. Permission prompts are useful, but they are not a full governance strategy by themselves. 

The company also needs monitoring after deployment. Permissions should not be checked once and forgotten. They should be reviewed continuously. Access rights change. Employees change roles. Vendors are added. Shared folders grow. Sensitive documents get copied. AI systems may be connected to new applications over time. Without ongoing review, the original rollout plan can become outdated quickly.
Liability is another reason to proceed carefully.

The Air Canada chatbot case is a useful warning. In that matter, a customer received incorrect bereavement fare information from Air Canada’s chatbot. The British Columbia Civil Resolution Tribunal found Air Canada responsible for the information provided through its chatbot and awarded damages. The broader lesson is straightforward. A company may not be able to avoid responsibility by saying the AI gave the wrong answer. 
There have also been widely discussed incidents involving customer facing bots making absurd statements, including a Chevrolet dealership chatbot that was manipulated into discussing a vehicle sale for $1. That incident appears to be more of a cautionary example than a binding legal precedent, but it illustrates the business problem. If AI is allowed to interact with customers without tight boundaries, the company may face reputational, contractual, regulatory, or customer service fallout.

Regulated companies also have to think about how they describe AI externally. FINRA reminded member firms in 2024 that the use of Gen AI and large language models does not remove existing regulatory obligations. The SEC has also brought AI washing cases, including settled charges against two investment advisers that agreed to pay $400,000 in combined civil penalties over allegedly false and misleading AI claims. 

In other words, companies need governance over both AI usage and AI marketing.

Insurance should be part of the planning process as well. This is an area many companies overlook. Cyber insurance, E&O, D&O, employment practices liability, media liability, and professional liability policies may respond to some AI related issues, but coverage will depend on the facts, policy wording, exclusions, and how the AI tool was used.

The insurance market is also evolving. The Financial Times reported that insurers at Lloyd’s introduced coverage aimed at losses caused by AI chatbot errors, including legal costs and damages tied to underperforming AI tools. That does not mean every company can easily transfer AI risk to an insurer, or that every claim will be covered. It does mean companies should talk with their broker and counsel before deployment, not after a problem occurs. 

Questions to ask include: does our cyber policy address AI caused data exposure? Does E&O cover bad advice or customer harm involving AI output? Does D&O respond to oversight claims involving AI governance? Does employment practices coverage apply if AI is used in HR workflows? Are AI vendors required to carry insurance? Do vendor contracts include indemnity? Are there exclusions for automated decisions, privacy violations, unauthorized system changes, or professional services?

The practical path forward is not complicated, but it does require discipline.

Start by inventorying sensitive data. Fix permissions. Classify information by risk. Choose enterprise tools with admin controls. Begin with lower risk, read only use cases. Use isolated environments for agents. Require approval for write, delete, send, publish, and external sharing actions.

Back up systems frequently. Monitor for unauthorized changes. Keep records of prompts, outputs, approvals, and actions. Train employees on what they should not upload. Review vendor contracts and insurance coverage. Then keep checking.
AI can be a meaningful tool for regulated companies. It can help employees move faster, reduce repetitive work, improve access to knowledge, and support better internal processes. But the companies that benefit most are likely to be the ones that treat AI deployment as a governance project, not just a software rollout.

Start small. Protect the data. Keep humans in the loop. Monitor continuously. Then expand with confidence, one controlled use case at a time.

This information is provided for general business and educational purposes only. Apex Technology Services provides cybersecurity, IT, and AI readiness guidance, but this content should not be considered legal, regulatory, or insurance advice. Organizations should consult qualified legal counsel, compliance professionals, and insurance advisors regarding their specific obligations, coverage, and risk exposure before deploying AI systems.

Aside from his role as CEO of Apex Technology Services and CEO of TMC, Rich Tehrani is CEO of RT Advisors and a Registered Representative (investment banker) with and offering securities through Four Points Capital Partners LLC (Four Points) (Member FINRA/SIPC). He handles capital/debt raises as well as M&A. RT Advisors is not owned by Four Points.

The above is not an endorsement or recommendation to buy/sell any security or sector mentioned. No companies mentioned above are current or past clients of RT Advisors.

The views and opinions expressed above are those of the participants. While believed to be reliable, the information has not been independently verified for accuracy. Any broad, general statements made herein are provided for context only and should not be construed as exhaustive or universally applicable.

Portions of this article may have been developed with the assistance of artificial intelligence, which may have contributed to ideation, content generation, factual review, or editing