Skip to content
Post Mortem

Real incidents, explained so anyone can follow

Each case starts with a plain-English explanation and a glossary, then covers the facts, the timeline, and what you can do today. Every claim links to its source, and we say clearly when something is analysis.

Reported incidents

Mid-October 2026 (reported October 19)
HIGH

OpenCode: exposed keys and unauthorized AI charges

Attackers got into an older OpenCode system and used stored API keys to run up charges on people's AI accounts.
October 21 to 26, 2026
CRITICAL

PHP "php2shell": critical flaws that let strangers run code

Two serious flaws in PHP, software that runs a large share of websites, let an outsider take control of a server. Fixes shipped, and CISA warned they were being exploited.
November 9, 2026
WATCH

Apple Bluetooth flaws: when "nearby" is the attack

Apple fixed several Bluetooth flaws, including ones that could let someone nearby pair with a device without the owner doing anything.
July 2025
HIGH

Replit's AI agent deletes a live database during a code freeze

A founder testing an AI coding tool said its agent wiped a live company database even though he had told it not to change anything.
July 2025
HIGH

Amazon Q coding extension shipped with a destructive prompt

An attacker got malicious instructions into a popular AI coding extension. AWS said the code shipped but did not run.
August 26 to 27, 2025
CRITICAL

Nx "s1ngularity": malware that borrowed your AI tools

Poisoned versions of a popular developer package stole passwords and tokens, and in some cases used installed AI command-line tools to hunt for them.
August 2025
CRITICAL

Salesloft Drift: stolen app tokens open hundreds of accounts

Attackers stole login tokens belonging to a connected chat app, then used them to quietly download data from customers' Salesforce systems.
CVE published May 2025
HIGH

AI-built apps with their database door left open (Lovable and Supabase)

A researcher found that many apps generated with an AI app builder let anyone read or write their database, because one switch was missing. The tool's maker disputes the CVE, saying app owners are responsible for their data.
Discovered January 31, 2026
HIGH

Moltbook: a viral AI-agent network with an open database

Researchers found a vibe-coded social network for AI agents had a misconfigured database that exposed 1.5 million API tokens and 35,000 email addresses.
May 2025
HIGH

GitHub MCP: a public issue steals private code

Researchers showed that a malicious comment on a public project could trick a connected AI agent into leaking private repository data.

Go further

Practice example