Shadow AI and AI data leakage

Your people already use AI at work, often on personal accounts. Above shows which tools, which accounts and which data, coaches people toward the approved tool at the moment they reach for another one, and investigates the patterns coaching doesn't fix.

A river splits into channels that carry paper boats across a plain. One channel runs off the land's edge with a violet boat on it.

INC-022A strategy director

Coached, closed

Board secrets go into a free AI slide tool

  1. Google SlidesBoard deck uploaded
  2. Google DriveFY27 plan pasted in
  3. Google SheetsM&A target list uploaded

Response Coach the director and three teammates

AI at work, by the numbers

  • 45%

    of employees are regular AI users on corporate devices, up from 15%. Verizon 2026 Data Breach Investigations Report.

  • 67%

    of users reach AI services through non-corporate accounts on corporate devices. Verizon 2026 Data Breach Investigations Report.

  • 43%

    of organizations had incidents involving unapproved AI, up from 20%. IBM, Cost of a Data Breach 2026.

Why shadow AI is hard to govern

“If we ban it, they use their phones”

When the approved tool is slower or missing, people open a personal account on the same laptop, or on a device you never see. A ban hides the usage. It doesn't end it.

“It's the same site with a different login”

ChatGPT on the corporate workspace is approved. The same site on a personal account is not. Rules written for domains can't tell the two apart.

“A custom GPT keeps everything it's given”

One paste is a leak. Thirty files uploaded into a personal project or custom GPT is a working copy of company material that stays with the person, wherever they go next.

The prettier deck

The prettier deck

Above investigative agents

  • Timeline
  • Contextual analysis
  • Reasoning
  • Recommended actions

Ready for security, HR, and legal to act on together.

What the investigation hands you

Ready for security, HR, and legal to act on together.

The prettier deckINC-022 · A strategy director

Coached, closed

Board secrets go into a free AI slide tool

Every step was an ordinary upload or sign-up to a web app, so DLP and CASB stayed quiet.

The evidence

  • Google Slides
  • Google Drive
  • Google Sheets
  • Gamma
  • Claude

Three uploads, the share with the team, two more sign-ups and the approved workspace nobody used.

The response

  1. Coach the director and three teammates to the approved tool
  2. Ask the vendor to delete the uploads
  3. Flag the tool

A human confirms

Talk to a human, 30 minutes

A solutions engineer walks you through a real case.

How Above handles shadow AI

  1. Tools, accounts, teams and data, from the browser and Anthropic's Compliance API.

  2. The approved tool and the reason, shown right there. Blocking is optional.

  3. Repeat uploads or AI use near a resignation become a case.

Watch the AI notetaker case
  • ChatGPT
  • Claude
  • Microsoft Purview
  • Google Drive
  • Slack
  • Okta

Evidence this investigation reads

The signals Above reads for shadow AI

They guide which AI tools to approve, and find misplaced data.

Show the signals from
  • Personal accounts on AI tools

    ChatGPT, Claude

    Corporate or personal, seen by the browser extension.

  • Sensitive data in prompts

    Microsoft Purview

    Customer data, source code or labeled files.

  • Uploads into custom GPTs and projects

    ChatGPT, Claude

    Read as a running total.

  • AI tools outside the approved list

    New AI sites, notetakers and extensions nobody reviewed.

  • Usage by team and tool

    Who uses which tool, against your AI policy.

  • Repeats after coaching

    The same paste after the approved path was shown.

  • Files pulled from shared drives into AI

    Google Drive

    Downloaded, then uploaded to an AI tool minutes later.

  • Corporate AI conversations

    Claude

    Claude Enterprise, through the Compliance API.

  • AI use around an HR event

    Workday

    Heavy personal AI uploads before a resignation.

Built to say yes to AI

Shadow AI is mostly people trying to work faster. Above starts with visibility and coaching, keeps investigations for the patterns coaching does not fix, and treats most cases as a reason to teach.

  • Coaching names the approved tool and the reason, and people can give a business reason.
  • Prompts stay evidence inside a case, masked and behind role-based access.
  • Keystroke logging is set by your policy, and there is no model training on your data.

Related use cases

AI agent insider threat

What AI agents did with the access they were given, who they acted for, and whether it matched the ask.

  • Departing employee data theft

    Customer lists, pricing and code that leave in the weeks around a resignation.

  • Shadow IT and unapproved software

    Software, extensions and SaaS accounts IT never approved, and which ones carry risk.

Questions about shadow AI

What is shadow AI?

Shadow AI is AI used at work outside the tools and accounts the company approved: a personal ChatGPT or Claude account, an AI notetaker, an AI browser extension, or a custom GPT built from company files. The Insider Threat Matrix™ lists exfiltration through a generative AI platform as IF001.006.

Should we block ChatGPT and other AI tools?

Blocking usually moves the usage to phones and personal laptops, where you see nothing. Most teams get further by approving a corporate tool, showing people that tool when they reach for another, and investigating the patterns that coaching doesn't fix. Above supports blocking as a policy option where you need it.

How is Above different from a CASB or SSPM for shadow AI?

A CASB or SSPM finds which AI and SaaS apps are in use and enforces rules on them. Above works beside them at the level of the person: which account was signed in, what data went in and whether it fits the job. It coaches in the moment and feeds policy back into Zscaler, Netskope or Microsoft Purview.

Can Above tell a personal ChatGPT account from our corporate workspace?

Above tells a personal ChatGPT account from your corporate workspace by the account that is signed in. The browser extension sees which account sends each prompt or upload, so the same site can be approved on the corporate workspace and coached on a personal account, without a blanket rule for the whole domain.

Are custom GPTs and AI projects a data leakage risk?

They can be. A custom GPT or a Claude project keeps the files it was given, so dozens of uploads into a personal one add up to a working copy of company material outside your control. Above reads the uploads, the project and its name as one pattern, and treats a single paste differently from a month of uploads.

Does Above read AI prompts?

Above reads AI prompts where you enable it. The browser extension captures prompts sent from managed browsers, and Claude Enterprise conversations arrive through Anthropic's Compliance API. They are kept as evidence for investigations, masked and access-controlled, under retention you configure, and never used to train models.

What security software can investigate shadow AI and unauthorized SaaS apps?

Shadow AI and unauthorized SaaS apps are covered by three kinds of software. Network tools such as Microsoft Entra Global Secure Access discover which AI apps people reach. CASB and SSPM tools inventory SaaS apps and their permissions. Above investigates the person: which account was signed in, what data went in, and whether it fits the job.

How do you catch employees connecting unapproved AI tools to company data?

You catch employees connecting unapproved AI tools to company data by reading the browser, the identity provider and the data together. The Above browser extension sees AI sites, personal accounts and extensions, the identity provider shows apps granted access to work accounts, and Above coaches people toward the approved tool at the moment they reach for another.

See where AI is already in use

See AI usage, coaching and investigations in one place.

Watch an investigation