Data-in-use risk is shifting insider security beyond documents

At a glance

What this is: This analysis says insider risk is moving from document exfiltration to data-in-use inside everyday collaboration channels, with AI prompts, screenshots and pasted text now central to the control problem.

Why it matters: IAM, PAM, NHI and human identity programmes have to account for how people and systems move data across web, email, chat and AI tools, not just how files leave the endpoint.

Context

The governance gap here is not that organisations lack policy, but that their policies were built around files, endpoints and obvious exfiltration patterns. Modern work now pushes sensitive information through collaboration tools, web apps and AI interfaces before it ever becomes a formal document, which makes detection and prevention much harder.

That shift matters because insider risk is increasingly behavioural and contextual rather than exceptional. When ordinary work patterns create exposure, security teams need controls that understand role, channel and data form at the point of use, including the way humans interact with AI systems and unmanaged sharing surfaces.

Safetica’s data is therefore less a story about a single tool class than about the widening gap between static policy and how data actually moves in day-to-day work.

Key questions

Q: What breaks when insider controls still focus on files instead of data in use?

A: Controls miss the most common modern leakage paths because sensitive content is increasingly copied, pasted, screenshotted or entered into collaboration tools before it becomes a file. That leaves policy blind to routine work behaviour and encourages workaround channels. Security teams need inspection and enforcement at the moment data is handled, not only when it is stored or transferred.

Q: Why does data-in-use risk rise when employees use collaboration tools and AI apps?

A: Because those tools sit inside everyday work, they become the natural place for people to move information quickly and informally. The problem is not only malicious intent. It is that normal work patterns can expose sensitive data when the control model assumes static documents instead of live interactions. Governance has to reflect how work actually happens.

Q: What signs show that insider policy is too rigid or too narrow?

A: Look for repeated blocks in ordinary channels, rising use of screenshots or pasted text, and users shifting to different apps after enforcement triggers. Those patterns usually mean the control is catching behaviour too late or in the wrong place. A healthy programme should reduce risky movement without simply displacing it into another channel.

Q: How should security teams govern sensitive data used by AI systems?

A: Security teams should treat AI as a data consumer that needs policy boundaries, not just authentication. Classify sensitive data, define which datasets may enter AI workflows, and monitor outputs, logs, and downstream reuse. If governance stops at login, the organisation can approve access while still losing control of the data itself.

Technical breakdown

Why data-in-use is harder to govern than document loss

Data-in-use is information being copied, pasted, screenshotted or entered into a service while it is actively being handled, not stored as a neat file. Traditional DLP and insider controls were designed around files, email attachments and obvious transfer events. That leaves a blind spot when sensitive content lives briefly inside chat, browser sessions, AI prompts or screenshots, because the data may never traverse the control point those policies inspect.

Practical implication: teams need controls that inspect content at interaction time, not only at file egress.

How collaboration channels become the real risk surface

Web apps, email and instant messaging concentrate routine business activity, so they also concentrate accidental leakage, policy violations and sanctioned-but-risky sharing. The issue is not that these channels are inherently malicious. It is that policy often treats them as generic transport rather than context-rich work surfaces where role, content type and intent vary from message to message. That makes blanket blocking noisy and easy to work around.

Practical implication: build channel-specific controls that distinguish approved collaboration from risky transfer patterns.

Why AI inputs change insider governance

Generative AI introduces a new data-handling pattern: users paste operational context, customer data or code into prompts that may never be stored in a managed repository. That turns prompts, screenshots and free-form text into governance objects, even when they are not formal records. For identity teams, this creates a policy problem as well as a usage problem, because access rights alone do not explain how a user decides to externalise data through an AI session.

Practical implication: treat AI tool access as a governed data pathway and apply policy to prompt content and session behaviour.

Threat narrative

Attacker objective: The objective is to move sensitive information out of governed workflows without triggering the controls built for files and formal transfers.

1. Entry begins with ordinary work in collaboration tools, browsers or messaging apps where sensitive data is already in motion.
2. Credentialed users copy, paste, screenshot or forward material into channels that were not anticipated by file-centric policy.
3. The impact is uncontrolled data exposure through routine business activity, including AI tools and unmanaged sharing paths.

Breaches seen in the wild

• Spain's first AI agent data breach 2026: Spain's AEPD logged its first breach notification attributed to an attacker's AI agent, which altered personal data and accessed invoices.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

File-centric insider control is now a governance lag, not a governance strategy. The core problem is that security teams still model leakage as document movement, while actual exposure increasingly happens in chat, browser flows, screenshots and AI prompts. That creates a control gap between where policy is written and where work is performed. The practitioner conclusion is straightforward: data governance has to follow the interaction, not just the repository.

Data-in-use risk is the named concept security leaders should adopt. It captures pasted text, screenshots and other transient content that never becomes a conventional file but still carries business sensitivity. That matters because most policy engines and review processes were designed to classify stored data, not live handling behaviour. The right governance question is whether the organisation can understand and control information at the moment it is being used.

Context-aware policy is now the minimum viable insider control model. The article shows that risk concentrates in ordinary work channels, which means role and context matter more than static allow or deny lists. Broad restrictions push users toward workarounds, while context-aware controls can preserve collaboration and still reduce exposure. The practitioner conclusion is to move from blanket blocking to governed, observable data handling.

AI governance and insider governance are converging at the prompt layer. Once employees paste operational data into GenAI tools, the distinction between user behaviour and data protection narrows sharply. That is not a separate policy problem. It is the same governance problem expressed through a new interface, and the conclusion is that AI usage must be monitored as part of the broader insider control plane.

Human identity controls increasingly depend on behaviour, not just entitlement. Access rights tell you who can enter a system, but not how they will move data once inside it. When normal users become the main source of risk, governance has to account for behavioural signals, workflow friction and channel switching. The practitioner conclusion is to align identity policy with observed usage patterns, not with role definitions alone.

What this signals

Data-in-use is the control gap many insider programmes are still missing, because the sensitive event happens during handling rather than after storage. That means identity and security teams need to think about behaviour at the point of interaction, especially where collaboration, AI and unmanaged sharing tools intersect.

The practical shift is toward context-aware governance. Static rules that assume documents are the primary unit of risk will keep generating noise, while channel-aware controls can preserve productivity and still surface the behaviours that matter most.

For practitioners

• Map your data-in-use surfaces Inventory where sensitive information is handled in practice, including browser sessions, chat, email, screenshots and AI prompts, then decide which surfaces need inspection and which need restriction.
• Tune policies to role and context Replace uniform rules with controls that consider user role, content type, destination channel and business context so legitimate collaboration does not trigger constant exceptions.
• Govern AI tools as data pathways Treat GenAI interfaces as controlled handling points for sensitive data, with explicit rules for pasted content, screenshots and unmanaged accounts.
• Measure channel switching after blocks Track whether blocked email, web or messaging activity reappears in another channel, because displacement often signals that policy is pushing risk sideways rather than reducing it.
• Review removable-media exceptions Check whether USB and other external transfer paths are still enabled for workflows that could be handled more securely through governed collaboration tools.

Key takeaways

• Insider risk is increasingly created by ordinary work patterns, not only by clearly malicious behaviour.
• Free-form text, screenshots and AI prompts now matter because sensitive information often never becomes a traditional file.
• Security teams need context-aware controls that govern how data is handled across collaboration and AI channels.

Key terms

• Data In Use: Data in use is information being actively accessed, processed, or modified by a user, application, or workload. It is the hardest state to govern because policy must follow live identity behaviour, not just storage location or network transit.
• Context-Aware Controls: Context-aware controls evaluate who is sending data, who should receive it, and what information is allowed in that interaction. They are important in support workflows because they reduce the risk of exposing the wrong customer’s data, leaking restricted content, or bypassing privacy rules through simple human error.
• Channel switching: Channel switching is the pattern where users move activity from one blocked or watched path to another, such as shifting from email to chat or from a file upload to a browser-based transfer. It is a useful signal that a control is displacing behaviour rather than reducing risk.
• Insider Risk Management: Insider Risk Management is the practice of detecting, investigating, and reducing harm caused by legitimate identities misusing access. It covers human error, malicious insiders, compromised accounts, and increasingly AI-driven actors that can move sensitive data without breaking perimeter controls.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, secrets management and workload identity. It is designed for practitioners who need to connect identity control to real operational risk.