The most significant development in insider risk during the first half of 2026 was the rapid spread of AI across everyday workplace tools. This created data-transfer channels that many organisations have yet to identify or control.
Some companies may believe they understand how employees use generative AI because major chatbot websites are restricted or monitored. In practice, AI is now embedded almost everywhere, including browser extensions, mobile and desktop applications, messengers, document and photo editors, and other business software.
Blocking a few well-known websites therefore provides only a partial view of where AI is being used and what information employees submit to it. Verizon’s 2026 Data Breach Investigations Report puts figures to this: 45% of employees are now regular AI users on corporate devices, roughly three times the share recorded a year earlier, and 67% of them sign in using non-corporate accounts. Shadow AI has become the third most common non-malicious insider action in the report’s DLP dataset.
When AI use does not feel like data transfer
Many employees do not view entering information into AI tools in the same way as sending it to an external recipient. Yet once corporate data is submitted to an external AI service, it has left the organisation’s controlled environment.
Platforms may offer controls to limit data retention or model training, but organisations cannot rely on every employee configuring those settings correctly. Moreover, even when all available restrictions are enabled, there is no guarantee that the AI system will not still gain access to the data through legitimate use.
AI services are also becoming indirect data transfer channels. An employee may upload documents from a work computer and later open the same conversation through a personal account on a home device. The platform then effectively functions as external storage.
We have encountered precisely this method in practice: An employee uploaded confidential corporate information from a work device and later accessed it outside the company through the same personal AI account.
Established risks have not disappeared
AI does not replace the traditional causes of insider incidents. Employees and contractors continue to gain access to more systems, cloud platforms, and collaboration tools. The same Verizon report found a third party involved in 48% of breaches, up from 30% the year before. In Malaysia, the cost is already documented: PIKOM’s “Beyond Compliance: The State of Cyber Resilience Malaysia 2026” report puts the average data breach at RM3.2 million, with some organisations reporting losses above RM5 million from a single incident, and 35.9% of the organisations surveyed experienced at least one cybersecurity incident between January 2024 and December 2025. The more data channels used in daily operations, the greater the opportunity for misuse of confidential information or legitimate privileges.
Compromised credentials create an additional risk. Criminals can operate through a legitimate employee account, making their actions initially appear to come from a trusted user. AI-generated phishing, voice imitation, and executive impersonation make such attacks more convincing. In the PIKOM data, AI-generated phishing and deepfake impersonation are already the most common attack type reported by Malaysian organisations at 32.6%, ahead of malware and ransomware-as-a-service at 30.2% and credential theft at 25.6%.
Corporate fraud also remains a major part of the insider-risk landscape. It may involve data theft, document forgery, conflicts of interest, misuse of company property, or the exploitation of internal information for personal gain.
Building visibility and control
Organisations should begin by establishing visibility over sensitive information. They need to know what data they hold, where it is stored, how it is classified, and who can access it. File repositories and cloud storage should be audited, excessive permissions removed, and changes to access rights monitored. DCAP tools can provide this visibility.
Second, companies need to control how data moves through all available transfer channels: email, cloud services, messengers, collaboration platforms, removable devices, printers, remote-access tools, and AI-enabled applications. Controls must follow the data rather than focus on a short list of applications. DLP systems help detect and prevent risky transfers across these channels.
Companies should also analyse behavioural and data-activity anomalies, including unusual uploads, changes in file activity, and deviations from normal working patterns.
Employee training remains essential, but generic warnings about “responsible AI use” are no longer enough. Organisations need clear, enforceable rules covering approved tools, prohibited data, and reporting procedures. Staff should also be trained to recognise phishing, impersonation, and other social-engineering attempts involving sensitive data, credentials, or payments.
