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)
HIGHOpenCode: 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
CRITICALPHP "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
WATCHApple 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
HIGHReplit'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
HIGHAmazon 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
CRITICALNx "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
CRITICALSalesloft 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
HIGHAI-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
HIGHMoltbook: 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
HIGHGitHub 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
32 cases 路 2007 to 2025
The biggest cyber attacks in history
M&S, Change Healthcare, MOVEit, 23andMe, Colonial Pipeline, WannaCry and more, newest first, with sources.
6 systems
Collapse scenarios
How governments and criminals get into water, power, fuel, banking, phones and hospitals, and what you would notice.