A Cursor AI agent autonomously deleted PocketOS's production database on April 25, 2026, after finding an over-scoped API credential; the incident is one of three in five weeks tracing to the same structural failure: AI agents creating long-lived, unowned credentials faster than governance can control them.

On April 25, 2026, a Cursor AI coding agent deleted the entire production database of PocketOS, a SaaS platform serving car rental businesses, in fewer than ten seconds, including volume-level backups. The agent was assigned a routine staging task, encountered a credential mismatch, autonomously scanned the codebase, and found an API token provisioned for domain management via the Railway CLI that carried blanket API authority across the entire Railway account.

GitGuardian's State of Secrets Sprawl 2026 report documented 28.65 million new hardcoded secrets exposed in public GitHub commits across 2025, a 34% year-over-year increase. AI-assisted commits leak secrets at roughly twice the GitHub-wide baseline. GitGuardian also found 24,008 unique secrets exposed in Model Context Protocol (MCP, the standard connecting AI agents to external tools) configuration files on public GitHub, with over 2,100 confirmed as valid live credentials.

Three distinct incidents in five weeks—the PocketOS deletion on April 25, a malicious LiteLLM package compromise on March 24, and a Vercel breach on April 19—all trace to credentials with excessive scope and no formal lifecycle governance. According to industry research cited at RSAC 2026, machine identities already outnumber human identities 45 to 1 at most enterprises; a Gravitee survey found that only 21.9% of teams have onboarded agent OAuth credentials into a privileged access management platform.