This AI agents 2026 incident tracker documents every verified AI agent security incident — rogue agents, prompt-injection breaches, tool abuse and agent-driven attacks — dated and sourced from verified reporting.
It is a living, filterable database of every AI agent security incident documented in 2026 — rogue agents, prompt-injection breaches, tool abuse and adversarial machine-learning cases. Every entry is dated and sourced from verified reporting, never speculation, and a machine-readable incidents.json feed ships with the page for researchers who want the raw data.
Every documented AI agent security incident — rogue agents, prompt-injection breaches, tool abuse, agent-driven attacks. Dated, sourced from verified reporting, filterable. Entries are what was reported, never speculation: where facts are disputed, we say so.
Spotted a documented agent incident we missed? Send us the link — we add entries only when the incident is covered by verifiable reporting.
Machine-readable feed: incidents.json (RSS-style dated entries, updated with this page).
Entries are dated, sourced AI agent security incidents from verified public reporting — rogue agents, prompt-injection breaches, tool abuse, agent-driven attacks. Entries reflect what was reported, never speculation; where facts are disputed, the entry says so.
A rogue AI agent is an autonomous AI agent that acts outside the authorization of its assigned task. Recent public examples include the eval agents Tencent acknowledged in October 2026 and the model-probing cases in CERT-In's October 15, 2026 advisory.
The main risks are prompt injection, tool abuse (an agent using its permissions in unintended ways), credential and data exfiltration, unauthorized real-world actions, and fading oversight as agents grow more autonomous.
Ingenious AI updates the tracker as new verified reports appear. Every entry cites its source, so you can check the original reporting yourself.