One attacker ran Claude Code and GPT-4.1 in parallel. Nine government agencies fell. 400 million records gone.

Check Point's 2026 AI Security Report dropped yesterday.

The finding that should alarm every CISO:

AI has crossed from assistant to operator.

One attacker breached nine Mexican government agencies using two AI models in parallel. Claude Code handled live exploitation across 34 sessions. GPT-4.1 analyzed stolen data and auto-tasked follow-on activity.

5,317 commands executed. 400 million records exposed. Minimal human direction between steps.

This is not a proof of concept. This is a production breach.

The vulnerability window just collapsed. AI turns a fresh disclosure into a working exploit in hours, not days. CISA now mandates 3-day remediation for critical flaws. India's CERT-In went further: 12 hours for internet-facing systems.

Your patch cadence is now a competitive disadvantage if it measured in weeks.

Meanwhile, your own AI stack is leaking. Check Point recorded a 5x increase in malicious prompt injection payloads between March and May. High-risk enterprise AI prompts doubled year-over-year. 87-93% of organizations saw at least one high-risk AI interaction last month. The average company runs 10 AI apps, most without formal approval.

Your employees are exposing credentials and source code through ordinary, approved AI use. No attack required.

Audit your AI attack surface today. Not next quarter. Today.

SOURCE: https://blog.checkpoint.com/ai-security/ai-security-threats-in-2026-insights-from-check-point-research/
VERIFIED: Check Point Research blog (July 13), PR Newswire press release (July 14), Help Net Security (July 15)
SIGNAL: The era of autonomous AI cyberattacks is here. Every enterprise running AI is now both a user and a target. Your security model must change or you will be the next nine agencies.