shadow AI insider threats data breach ransomware AI security

Shadow AI Is the Insider Threat Nobody's Watching

21 May 2026

Verizon's 2026 Data Breach Investigations Report reveals a fourfold increase in "shadow AI" use, with 67% of employees who regularly use AI at work doing so through unauthorized personal accounts, potentially exposing sensitive corporate data such as source code, documents, and proprietary research to unvetted third-party platforms. The report also highlights worsening vulnerability management, with remediation rates for critical flaws dropping from 38% to 26% and resolution times rising from 32 to 43 days, while ransomware featured in nearly half of all breaches. On a positive note, ransom payments continued to decline, with 69% of victims refusing to pay and the median payment falling slightly to just under $140,000.

You've heard of shadow IT. Shadow AI is here now, and it's growing fast.

Verizon's latest Data Breach Investigations Report, covering more than 22,000 incidents globally, has put some numbers to what many security teams already suspected: employees are routinely feeding sensitive corporate data into AI tools their IT departments have never sanctioned. The scale is the alarming part. Among the 45 percent of professionals who use AI regularly at work, a full 67 percent are doing so through personal accounts. That's not just ChatGPT. It covers coding assistants, AI agents, external chatbots, and whatever vibe coding platform your developers have discovered this week.

The result is a fourfold increase in non-malicious insider incidents compared to last year's figures. Non-malicious is doing a lot of work in that sentence. Intent doesn't matter much when source code, proprietary research, internal documents, and structured data are all leaving the building via a third-party model nobody vetted.

Verizon found that 28 percent of data loss prevention violations involved employees pasting source code into an AI tool. Images, structured datasets, documents and PDFs followed behind. In 3.2 percent of cases, workers were uploading proprietary research and technical documentation. That's a specific, awkward number. Not huge in percentage terms, but pointed.

The fix, according to Verizon, involves tightening enterprise configurations, locking down account permissions, and stopping users from blindly trusting whatever AI wrapper sits between their data and the underlying model. Easier said than done when half the workforce treats AI tools like a search engine with better grammar.

One response to all this is the AI Bill of Materials, or AI-BOM. Cisco open-sourced its version earlier this year and has since released tooling to track model provenance. Palo Alto Networks' Ian Swanson made the case recently that AI-BOMs aren't just useful for compliance box-ticking. They give incident responders a way to reconstruct what an AI system looked like at a specific point in time, which becomes rather important when an attacker has quietly modified a system prompt to make your own AI work against you. 'What system prompt was used within the ingredients to create the AI application? And then see it's changed from a prior state to a new state,' Swanson said. That kind of forensic capability is going to matter more as AI gets woven deeper into enterprise infrastructure.

On the vulnerability front, exploitation of software flaws has overtaken credential abuse as the leading cause of breaches. Credential compromise is actually down 13 percent year-on-year, which counts as good news. The patching picture, however, is grim. Only 26 percent of critical vulnerabilities in CISA's Known Exploited Vulnerabilities catalogue were fully remediated in 2025, down from 38 percent the previous year. Median time to full resolution crept from 32 days to 43 days. To be fair, the number of critical vulnerabilities requiring attention jumped 50 percent, so security teams aren't slacking so much as drowning.

Ransomware featured in 48 percent of all breaches covered, up slightly from 44 percent. The one genuinely encouraging data point: ransom payments are declining. Sixty-nine percent of victims refused to pay, and the median payment dropped from $150,000 to just under $140,000. Small mercies.