In late November 2010, a cache of roughly a quarter of a million United States diplomatic cables began appearing in newspapers around the world. The BBC noted at the time that only about six percent of the documents were classified "secret" — the majority were unclassified or at the lower "confidential" level.[1] That detail is the whole story for anyone responsible for corporate information security. This was not a sophisticated intrusion against the most heavily protected material in government. It was a large volume of routine, moderately sensitive information, accessible to a very large population of cleared users, extracted by one of them. The cables had been held on SIPRNet, a classified network built after earlier intelligence failures specifically to widen information sharing across agencies. The analyst later convicted of the disclosure held a security clearance appropriate to the material, and the exfiltration reportedly involved little more than writing files to removable media while appearing to listen to music.[2] Every element of that sentence has a direct corporate equivalent.
The Sharing Paradox
SIPRNet existed because siloed information had been identified as a cause of failure. The remedy — broad access for cleared personnel — worked as intended and created a new exposure: a single authorised user could reach far more than their role required. Enterprises made the same trade-off in the same period and for the same reasons. Shared drives were opened up because access requests were slowing people down. Data warehouses were built to let analysts self-serve. Collaboration platforms were deployed precisely to remove barriers between teams. Each decision improved productivity and enlarged the blast radius of any single compromised or malicious account. The uncomfortable conclusion from 2010 is that the sharing was not the mistake. The mistake was sharing without detection — giving broad read access while retaining no capability to notice when someone used all of it at once.
Why Volume Is the Signal
A legitimate user reading documents relevant to their work produces an access pattern. That pattern has a shape: a certain number of files, from a certain set of locations, at a certain rate, during certain hours. Bulk exfiltration breaks every dimension of that pattern simultaneously. Volume far above the individual's baseline. Breadth across repositories the person has never touched. Sequential rather than selective access. Often unusual timing. And frequently a destination — removable media, personal cloud storage, a private email address — that has no business justification. None of those signals requires advanced analytics to detect. They require somebody to be looking, which in 2010 almost nobody was. Monitoring effort was directed outward at the perimeter, and the authenticated insider was treated as a solved problem because authentication had succeeded.
The Corporate Versions of This Scenario
The pattern repeats across industries with dispiriting regularity. A departing salesperson downloads the complete customer relationship management database in their final fortnight. An engineer copies source code repositories to personal storage before joining a competitor. A finance analyst extracts the full payroll file. A contractor with temporary elevated access takes the design documentation for a project they were briefly attached to. In each case, access was legitimate. Authentication succeeded. No control was bypassed. The only distinguishing feature was volume and purpose — and purpose is not visible to a permissions system. The scenario that is materially worse than 2010 is that the destination options have multiplied. The analyst in 2010 needed physical media. Today the same volume moves through a browser upload, a synced personal cloud folder, a messaging application, or a pasted extract into a consumer AI tool, in seconds, from a laptop at home.
What Actually Reduces This Risk
- Right-size access continuously, not at onboarding. Access accumulates as people change roles and nobody removes the old entitlements. Periodic recertification, with managers confirming each entitlement is still needed, removes more risk than any monitoring tool.
- Monitor for volume anomalies at the data layer. Downloads, exports, query result sizes and repository access counts, baselined per user. The absolute number matters less than the deviation from that individual's normal behaviour.
- Control the egress paths deliberately. Removable media, personal cloud storage, webmail, messaging uploads and AI tools. Blocking everything fails; knowing which paths exist, which are permitted and which are logged is achievable.
- Watch the leaver window closely. The period between resignation and departure concentrates a large share of insider incidents. Heightened monitoring during notice periods is proportionate and effective.
- Classify enough to prioritise. A full classification programme rarely finishes. Identifying the handful of repositories that would genuinely damage the business, and monitoring those properly, is achievable in a quarter.
- Separate bulk export rights from read rights. Many users need to read records. Very few need to export ten thousand of them. Treating export as a distinct, logged and limited privilege is one of the highest-value controls available.
- Log access to sensitive repositories and review the logs. Logs nobody reads are a compliance artefact, not a control. A short weekly review of anomalies beats a comprehensive archive nobody opens.
- Make the policy explicit and known. People behave differently when they know access to sensitive material is recorded and reviewed. Deterrence is cheap and underused.
| Review area | Evidence to establish |
|---|---|
| Entitlements | Which rights remain necessary after a role change? |
| Bulk export | Who may export records rather than read them? |
| Access behaviour | What departures in volume, breadth or timing are reviewed? |
| Egress | Which transfer paths are permitted and logged? |
| Repository priority | Which collections need the closest attention? |
| Review ownership | Who reviews anomalies and applies the stated policy? |
Qualitative summary of this article's source text, not a measured outcome or performance estimate.
The Motivation Question
Corporate insider risk programmes tend to assume financial motive, because that is what fraud controls are designed for. The 2010 case involved conviction rather than profit, and the disclosures that followed over the next fifteen years — across government and the private sector — have consistently included people acting from grievance, ideology or a belief that disclosure was justified. This matters operationally for one reason. A control set tuned to detect theft for financial gain looks for the patterns of theft for financial gain. Someone who believes they are doing the right thing does not conceal in the same way, does not follow the same timeline, and is frequently a trusted long-tenured employee rather than a recent joiner. Behaviour-based detection catches both; motive-based profiling catches neither reliably.
The Current Shape of the Problem
The insider exfiltration scenario has been altered by two developments. The first is that most corporate data now sits in cloud platforms with generous native export capabilities. The bulk download is a feature, documented in the help centre, available to anyone with read access. The second is AI. An employee pasting sensitive material into a consumer AI tool is committing an unintentional version of the same act — large-volume transfer of internal information to an external system with no contract, no retention control and no audit trail. And an AI assistant with standing access across a workspace can retrieve and summarise at a scale no human could, which means that a compromised assistant session is functionally a bulk exfiltration event with none of the traditional signals. The defensive principle established in 2010 still holds and now needs applying to non-human actors too. Authentication tells you who is asking. Only monitoring tells you whether what they are taking makes any sense.
Common Questions
What was the main security lesson of the 2010 diplomatic cable disclosures?
That broad legitimate access without volume monitoring is the core insider risk. Most of the material was not highly classified; the failure was that one authorised user could reach and extract an enormous quantity of it undetected.
How do you detect insider data exfiltration?
By baselining normal access behaviour per user and alerting on deviations in volume, breadth across repositories, access sequence, timing, and use of egress paths such as removable media, personal cloud storage or webmail.
When are insider incidents most likely?
Disproportionately during the notice period between resignation and departure, and among contractors or staff whose access was elevated temporarily and never reduced.
How does AI change insider risk?
It adds a high-volume, low-friction egress path through consumer AI tools, and introduces assistants with standing broad access whose retrieval activity can amount to bulk extraction without triggering traditional download-based detection.
Insider Risk Review — Outpace maps who can reach your most sensitive data, how much of it they could take, and which egress paths would notice — then closes the gap between authentication and detection.