← Back to the journal
AI & Governance

Does Your Organization Have an AI Policy?

Your employees adopted artificial intelligence long before your company wrote a single rule about it. The bill for that gap is starting to arrive.

A diverse team of young professionals working on MacBook laptops against a bold pink and red wall

Somewhere in your organization this afternoon, a sales manager is pasting a client contract into a chatbot and asking for the key terms. A finance analyst is uploading a quarterly spreadsheet to build a chart for Monday’s meeting. None of them asked permission. Most of them would be surprised to learn they needed to.

This is what security professionals call shadow AI: the use of AI tools at work without the knowledge or approval of the people responsible for protecting company data. A year ago it looked like a fringe behavior. It no longer does.

The scale of it

It is already most of your team

Measured from network traffic and large worker surveys, not wishful self-reporting.

45%Of employees now use AI regularly on corporate devices, up from 15% a year earlier (Verizon)
2 in 3Of those users reach AI through personal accounts their employer cannot see
88%Of employees given an approved AI tool also use a personal one for work (Gartner)
Verizon 2026 Data Breach Investigations Report; Gartner Q1 2026 Global Labor Market Survey.

The instinct of many leadership teams is to treat this as an IT matter. It is not. It is a question of governance, of brand, and increasingly of money. The organizations that get ahead of it will not be the ones that ban AI. They will be the ones that write down, clearly and early, how it should be used.

The leak nobody sees

The risk is mundane, which is precisely what makes it dangerous. When someone types or uploads information into an AI tool, that information leaves the building. Depending on the product, the plan and the settings, it may be stored, reviewed by the vendor, or used to train future models. On a personal account, the company has no record it happened at all.

The early warning came in 2023, when Samsung engineers pasted proprietary source code and internal meeting notes into ChatGPT, and the company restricted generative AI on its devices. At the time it read like a cautionary tale about one careless team. Verizon’s latest data suggests it was a preview. Shadow AI is now the third most common way employees accidentally expose company data, a fourfold jump in a year, and source code remains the most common thing they send.

The part with a dollar sign

The gap is now measurable

IBM studied 602 breached organizations for its 2026 Cost of a Data Breach Report.

43%Of security incidents involved shadow AI, up from 20% the year before
$5.39MAverage cost of an AI-related incident, against a $4.99M global average
68%Of organizations hit by an AI incident had no AI governance policy at all
IBM 2026 Cost of a Data Breach Report. 92% also lacked proper access controls for AI systems. The tools arrived. The rules did not.

What goes out matters as much as what goes in

Security conversations tend to focus on data flowing into AI. As a marketer, I worry just as much about what flows out and gets published under a company’s name.

Consider Deloitte Australia. In 2025 the firm agreed to partially refund the Australian government for a A$440,000 report after an academic found it riddled with errors, including references to work that did not exist and a fabricated quote attributed to a federal court judgment. The revised version disclosed that a generative AI model had been used in its preparation. A Big Four firm, a government client, a refund, and a news cycle no brand wants.

AI states invented facts with exactly the same confidence as real ones. The accountability stays with whoever hits publish.

The same applies to images and copy that drift too close to someone else’s copyrighted work, to drafts that quietly surface confidential details, and to content so generic it erodes the voice a brand spent years building. Disclosure is becoming a legal matter too: since August 2, 2026, the transparency rules in Article 50 of the EU AI Act have been enforceable, requiring organizations to tell people when they are interacting with an AI system and to label certain AI-generated content. Fines can reach €15 million or 3 percent of worldwide turnover. Any company with European customers, or a chatbot on its website, should already have an answer.

The dashboard that wasn’t meant to be public

I build dashboards with AI myself, and I understand the appeal. Tools such as Claude can turn a spreadsheet of key performance indicators into a polished, interactive dashboard in minutes: revenue by location, membership growth, cost per lead, staff performance. Work that once required an analyst and a week now takes a coffee break.

The catch is the share button. Many AI tools let users publish what they create through a link, and “anyone with the link” means exactly that. In late July 2026, users discovered that shared Claude conversations and Artifacts were turning up in Google search results, including business plans, resumes, API keys, and at least one dashboard apparently built to analyze medical billing data. No one was hacked. Someone pressed share to make life easier for a colleague, and a private link became a public page.

Before you publish an AI-built dashboard, ask

Does it contain financial, customer or employee data? Who exactly can open the link, and can it be forwarded? Was it built on a company account or a personal one? Who owns it, and can we take it down? And would we be comfortable if it landed on a competitor’s desk tomorrow?

Your KPIs tell the story of your business: where it is strong, where it is bleeding, which locations are underperforming. In the wrong hands, that is a competitor’s roadmap.

When AI stops answering and starts acting

The next stage of this problem is already here. AI tools no longer simply respond to prompts. They connect directly to email, cloud drives, calendars and CRMs, and increasingly they act as agents: sending messages, updating records and moving files on a user’s behalf.

That changes the calculus. A chatbot that receives a pasted paragraph sees one paragraph. An agent with access to a shared drive can see everything in it. BlackFog’s 2026 survey of 2,000 UK and US employees found that 51 percent had connected AI tools to other work systems without IT approval, and IBM found that fewer than half of organizations secure the credentials those AI workflows run on.

The principle that governs this is old and unglamorous: give any tool access only to what the task requires, review those connections regularly, and cut them off the day an employee leaves. Agents make the principle urgent, because an orphaned connection keeps working long after the person who set it up has gone.

What a workable policy looks like

None of this requires a 60-page legal document. The best AI policies I have seen fit on a single page and are written for the people who will follow them, not the lawyers who will file them. They tend to cover the same ground.

The data rules

Green, yellow, red

One simple way to classify what is allowed into an AI tool. Everyone can remember it.

Green

Public information and general research. Go ahead.

Yellow

Internal material. Allowed, but only in approved tools on business accounts.

Red

Customer data, employee records, financials, credentials, health, legal matters, anything under NDA. It never goes into AI.

Around those rules sit a handful of habits. Approved tools and accounts: name the tools your team may use and require business or enterprise plans, which typically commit not to train on your data. Personal accounts are never used for company information. Settings someone owns: data training, retention and admin controls configured deliberately by a named person, because defaults are not a policy. Human review before anything is published: every AI-assisted blog, post, ad and deliverable checked for accuracy, sourcing, confidentiality and voice, with every statistic verified. Rules for shared links, limits on connected apps and agents, a no-blame way to report mistakes, and training on a quarterly cadence, because the tools change faster than any annual review.

Start here

If you do nothing else this month

Six moves that take a week, not a quarter. Ask your team which AI tools they actually use, and expect to be surprised.

  • Ask your team which AI tools they actually use, and expect to be surprised
  • Move approved tools to business-level accounts
  • Share the green, yellow and red data rules
  • Audit what dashboards, chats and reports have already been shared by link
  • Require human review of all published AI content
  • Name one person who owns AI governance and put a quarterly review on the calendar
One page of rules beats sixty pages nobody reads.

The question for the boardroom

The companies that will struggle with AI over the next few years are not the ones that moved too slowly. They are the ones that let adoption run ahead of judgment, and discovered the gap only when a contract, a dashboard or a client report turned up somewhere it should not have been.

The fix is not complicated. It is a page of clear rules, a person who owns them, and a culture where employees ask before they paste. That is far cheaper than a breach, a fine, or a refund.

So the question is worth asking plainly, at the next leadership meeting: does your organization have an AI policy? If the honest answer is “not yet,” this quarter is the time to write one.

Ask before you paste.
Brea

Free download: AI Policy SOP Template, a standard operating procedure your organization can adapt

Take this further

Sources

BB
Brea Ballard

Founder and resident Iconist at House of Icon. Twenty-five years creating, building, launching, marketing, and growing brands. Still loud, still allergic to beige.

Share this

Keep reading

The House is open

Come hang
in the House.

Brand tips, behind-the-scenes chaos, and daily proof that bland is banned. Pick your poison.

@houseoficon.co