Briefing · 2026-06 · DRAFT, prose pending the editor
AI Failure Briefing: June 2026
The period in numbers
628 entries indexed in 2026-06 (25 the month before). The index stands at 670 live entries; 12 duplicates have been merged to date.
- Hallucination201 (+195)
- Policy Violation129 (+123)
- Brand & Safety Incident111 (+105)
- Agentic Action Error70 (+68)
- Tool Misuse37 (+35)
- Prompt Injection36 (+34)
- Data Leakage29 (+28)
- Identity & Access Drift15 (+15)
Severity mix this period: Catastrophic 5%, High 52%, Medium 37%, Low 6%.
The month in one paragraph
[Editor's note pending. One paragraph: what the month's intake says about where production AI is failing now.]
Five observations
- [Editor's note pending: trend observation grounded in the mode-share delta below.]
- [Editor's note pending: trend observation on the predictive wing.]
- [Editor's note pending: trend observation on severity mix.]
- [Editor's note pending: a notable single entry and why it matters.]
- [Editor's note pending: what to watch next month.]
Notable new entries
[Editor's note pending: three to five entries from this period worth a paragraph each.]
Highest-stakes entries this period
- FI-0027 · Catastrophic A Cursor AI agent deleted a startup's production database and backups in nine seconds (Identity & Access Drift)
- FI-0242 · Catastrophic OpenClaw ClawHub marketplace exploited to distribute macOS stealer malware (Tool Misuse)
- FI-0236 · Catastrophic Moonwell DeFi platform loses $1.78 million due to AI generated smart contract pricing error (Hallucination)
- FI-0262 · Catastrophic Brazilian firm allegedly used AI to illegally resell SUS patient data (Data Leakage)
- FI-0243 · Catastrophic OpenClaw agent skills suffer widespread vulnerabilities and data exfiltration (Prompt Injection)