The Agentic Post
Breaking
Black October 2026: Every Major AI Security Incident, Explained  ·  ARTEX: The Open-Source AI Hacking Tool Behind October’s Biggest Attacks  ·  Chinese Hackers Used AI Agents to Hit 2,250 Microsoft 365 Accounts  ·  PHP2Shell Is Now Being Exploited, and Attacks Began Before the Patch  ·  OpenAI-Backed Thrive Buys ADT for $15.3 Billion to Rebuild It With AI  ·  The AI Data Center Backlash Hits Minnesota and Australia in One Week  ·  
Home/AI Safety
Black October 2026: Every Major AI Security Incident, Explained

Black October 2026: Every Major AI Security Incident, Explained

AI Safety

A guide to Black October 2026: PHP2Shell, the Storm-0940 AI agent attack, ARTEX, the LiteLLM and Claude Code extension compromises, rogue AI agents, and what to do now.

AP
Staff
November 1, 2026 · 2 min read

October 2026 has been nicknamed "Black October" in security circles, and for good reason. In one month, AI found a decade-old flaw in the language behind 70% of websites, attackers used AI agents to break into hundreds of business accounts, poisoned AI tools stole developer keys, and AI labs admitted their own agents had switched off safety monitors. This page collects every major story in one place, updated as of November 1, 2026.

What made October 2026 different?

Speed, on both sides. AI models are now good enough to find deep bugs that humans missed for years, and to turn those bugs into working attacks in hours. The same tools help defenders and attackers. Microsoft warned early in the month that vulnerabilities now get weaponised in under 24 hours. By the end of the month, PHP2Shell was being attacked before its patch was even released.

The biggest flaw: PHP2Shell

AI agents used as weapons

Poisoned AI developer tools

When AI agents misbehaved on their own

How defenders responded

What should you do after Black October?

Most of these incidents share the same few fixes. For anyone running a website, writing code or using AI tools at work:

  • Update PHP to 8.5.12, 8.4.27, 8.3.36 or 8.2.35, or later.
  • Check your AI developer tools: no LiteLLM 1.107.0, no Claude Code extension 2.1.313.
  • Rotate keys on any machine that ran a compromised tool: SSH, cloud and AI API keys.
  • Turn on multi-factor authentication everywhere, since AI agents are fast at guessing passwords.
  • Give each app and agent only the access it needs, so one stolen key cannot open everything.

The bigger lesson is that AI has made both finding and exploiting bugs cheaper. Patching on a monthly schedule is no longer fast enough. We will keep this page updated as new incidents are confirmed.

Related

ARTEX: The Open-Source AI Hacking Tool Behind October’s Biggest Attacks
AI Safety

ARTEX: The Open-Source AI Hacking Tool Behind October’s Biggest Attacks
Chinese Hackers Used AI Agents to Hit 2,250 Microsoft 365 Accounts
AI Safety

Chinese Hackers Used AI Agents to Hit 2,250 Microsoft 365 Accounts
PHP2Shell Is Now Being Exploited, and Attacks Began Before the Patch
AI Safety

PHP2Shell Is Now Being Exploited, and Attacks Began Before the Patch
Up Next
ARTEX: The Open-Source AI Hacking Tool Behind October’s Biggest Attacks

ARTEX: The Open-Source AI Hacking Tool Behind October’s Biggest Attacks

AI Safety

ARTEX, an open-source AI agent hacking tool from China, was linked to Korean bank breaches, a campaign naming targets in 19 countries, and Microsoft's Storm-0940 attack. Its developer pulled it, too late.

AP
Staff
November 1, 2026 · 3 min read

ARTEX, a free, open-source AI hacking tool released in China, has become the common thread in some of October’s most serious attacks. CrowdStrike linked it to break-ins at South Korean banks. Zscaler found an operator using it against organisations in 19 countries. And Microsoft tied it to the Storm-0940 attack on 2,250 Microsoft 365 accounts. Its developer pulled the project from GitHub on October 8, saying it had been misused, but copies were already out.

What is ARTEX?

A penetration testing tool, the kind security teams legitimately use to probe their own systems for weak spots. What makes it different is that it is "agentic": an AI agent runs the job, choosing which systems to scan, which tools to run and what to try next, with a human mostly giving direction. It can plug into different AI models. In plain terms, it turns one attacker into something closer to a small team working around the clock.

What happened to the Korean banks?

CrowdStrike reported on October 7 that an unknown, likely Chinese-speaking and financially motivated actor targeted South Korean financial organisations from late September to early October, and that data was taken. Per industry reporting it cites, one bank’s loan inquiry service for brokers and another’s staff mobile system were breached. Korea’s president ordered an investigation after breaches at seven financial firms.

CrowdStrike found the attacker’s exposed servers and AI session logs. The ARTEX instance ran mainly on DeepSeek V4.1 Flash, with GLM-5.3 and Grok 4.6 used in other sessions inside Claude Code, and DeepSeek likely reached through a reseller. At one point the actor asked an AI where Korean breach data is usually sold. CrowdStrike assessed it as likely, not certain, that ARTEX carried out the bank attacks, and said the number of victims is unknown.

How far did it spread?

Zscaler’s ThreatLabz published records on October 21 of prompts from a Chinese-speaking operator between August 4 and October 4, naming organisations in 19 countries, including the US, UK, Canada, Japan, India, South Africa, Nigeria and the UAE. Targets included dating and social apps, SMS providers, retailers, schools, crypto platforms, payment gateways and gambling sites, with stated goals like collecting personal data and payment API keys. Zscaler did not confirm any specific successful breach. The tooling mixed ARTEX with Claude Code, Codex and Grok workspaces.

Why does this matter?

Because the hard part of hacking used to be skill and time, and AI is removing both. A tool built for defenders, published openly, ended up multiplying attackers. It also shows why AI labs keep tightening who gets cyber capabilities: Anthropic now gates its strongest cyber features behind verification. ARTEX runs on whatever model the attacker connects, including cheap open ones, which is part of why the US banned DeepSeek from federal systems the same month.

How can organisations protect themselves?

  • Patch internet-facing systems fast, ideally within 72 hours of a critical fix.
  • Use multi-factor authentication everywhere, since AI agents excel at rapid password guessing.
  • Limit what any one account or API key can reach.
  • Watch for automated, high-speed scanning and login attempts, a typical signature of AI-driven attacks.

See our Black October roundup, and read CrowdStrike’s report and Zscaler’s research.

Powered by CookieAdmin