Proofpoint warns AI will widen insider threat risks in 2026

What happened

Proofpoint warned in an online outlook published on 14 January 2026 that AI would expand insider threat risks for organisations, including through autonomous systems given access to confidential information.

The cybersecurity company’s blog, written by Kasey Olbrych, Stephanie Torto and Nirav Shah, sets out five predictions covering AI misuse, corporate upheaval, security monitoring, investigations and shared responsibility for managing risk. It is a forecast and guidance document, rather than a report of a newly discovered breach.

Its central argument is that insider risk programmes must account for software acting with trusted access, as well as the employees who use it. AI agents — systems able to carry out tasks across applications — can browse, write code and initiate actions without a person directing every step.

According to Proofpoint, agents may connect several tasks in ways that take them beyond their intended remit. Poor configuration can allow them to expose information or weaken safeguards, while hostile manipulation can steer their actions towards unauthorised outcomes.

The company also identifies risks from ordinary employee use of AI assistants. Workers may include confidential material in prompts, the instructions or questions submitted to an AI system, or use assistants to summarise information drawn from restricted sources.

Deliberate misuse presents a separate problem. Proofpoint says AI can provide technical guidance that lowers the expertise needed to abuse access, avoid monitoring or extract sensitive information. It also warns that confidential content can be reproduced in an AI-generated summary without the user directly copying the original file.

The background

Traditional insider threat programmes have focused on people with authorised access, examining their behaviour, motives and opportunities to cause harm. Proofpoint argues that AI changes both what those people can do and which systems should fall within the scope of security oversight.

This broader view also appears in DTEX’s guidance on including AI in insider risk management, which considers trusted AI systems alongside people, mistakes, deliberate misuse and stolen credentials. The common issue is how authorised access can lead to harm, rather than whether an attacker has broken into a network.

Proofpoint’s second prediction concerns organisational disruption. It expects recruitment competition, corporate espionage, mergers, acquisitions and business disposals to increase opportunities and incentives for insiders to take valuable information during 2026.

The company identifies customer lists, intellectual property and strategic intelligence as potential targets. It says employees may use sensitive information as leverage when changing jobs, while AI could help someone imitate legitimate requests or conceal suspicious activity.

Restructuring creates a different set of weaknesses. Transitional accounts, uncertain responsibility for systems and changing access arrangements can make it harder to establish who should retain permission to use particular resources, according to the outlook.

Although Proofpoint predicts an increase in incidents, the article does not quantify that expected rise or present a statistical model supporting it. Its reference to corporate espionage cases in 2025 provides background, not a measured basis for the scale of any increase in 2026.

What people are saying

Proofpoint says security teams should bring together employee behaviour, identity and employment information, and technical activity records. It argues that these sources provide a more useful picture when assessed together than when handled separately.

The company lists workplace friction, indications that someone is preparing to leave, financial pressure and possible coercion among the contextual signals it considers relevant. It also proposes examining employment circumstances such as leave status, performance processes and dissatisfaction around bonuses, alongside a person’s access rights.

Technical records provide another layer, according to the authors. Their examples include unusual access, misuse of elevated permissions, attempts to move data out of an organisation, and efforts to manipulate AI prompts or bypass controls. Proofpoint argues that combining these records with other context could reveal emerging risks weeks before information is lost, although the article supplies no measured results for that claim.

The outlook also describes AI as a tool for security teams. Proofpoint says it can connect related alerts, distinguish higher-priority incidents from less useful warnings and turn large volumes of activity records into readable investigation summaries. It cites Microsoft Security Copilot as an example of a generative AI tool that can produce summaries and suggest next steps.

The authors envisage agents gathering alerts, assembling incident chronologies and comparing user activity across systems, with human oversight. They also predict wider use of models that identify patterns associated with escalating risk and AI-generated response procedures that reduce manual investigation work.

What happens next

Proofpoint’s recommendations place responsibility across security, legal, compliance, privacy and human resources teams. It suggests establishing a joint insider and AI risk council, or extending an existing programme, to agree objectives, review incidents and maintain common standards.

That emphasis on shared oversight is consistent with Adaptive Security’s insider threat programme guidance, which combines technical monitoring with staff training and responsibility across departments. In Proofpoint’s outlook, the additional challenge is deciding who controls systems that can act on an organisation’s behalf.

The company recommends explicitly assigning authority to approve AI uses, define agent permissions, conduct ethics reviews and disable agents. These are distinct responsibilities: approving a business task does not, by itself, settle which information or applications an agent should be allowed to access.

It also calls for rules covering acceptable use, responsible AI, consent for data use and transparency. Proposed technical safeguards include permissions based on a person’s role and monitoring of user activity, intended to put those rules into practice rather than leave them as written policy alone.

The proposed use of employment information and communications makes privacy part of the design of the programme, not simply an issue after an alert. Proofpoint includes privacy specialists in its recommended oversight structure and says organisation-wide AI adoption requires clear privacy rules.

The article sets out priorities for insider risk teams entering 2026, but does not establish a timetable for implementation or a common standard against which progress will be measured. Its predictions remain prospective; the publication does not demonstrate that the expected changes have already occurred across organisations.