October 30, 20267 min read

Your API Key Just Went Shopping Without You: Inside Black October, the Month AI Started Picking the Internet’s Locks

What really happened in Black October, what the viral posts got wrong, why marketing owns more of the attack surface than it thinks, and an eight-step checklist.

Around 50 billing notifications. In one morning.

That’s how a startup founder found out his AI keys had been stolen. $281 to OpenAI. $168 to Anthropic. $449 of AI usage he never ran, billed to his card, by someone he will never meet (Fox News).

He didn’t click a dodgy link. He didn’t reuse a password. He simply used an AI coding platform that got breached.

Here’s the scary part: his story was one of the smaller ones from October 2026. The security world now has a name for the whole month. Black October.

If your first thought is “that’s an IT problem”, stay with me. Because a lot of the doors hackers tried this month were the website, the plugin or the AI tool that marketing set up.

So what actually happened?

Black October wasn’t one big hack. It was a pile-up on the highway, with every lane blocked at once.

Threat intelligence firm ScruteX tracked 28 separate incidents through the month, and ten of them surfaced in just 48 hours, on October 20 and 21 (ScruteX). Lumu described security teams applying patches almost non-stop as portals were forced offline (Lumu). Picture three kitchens on fire and one extinguisher.

Here’s the highlight reel.

1. PHP2Shell: a bug under most of the web

Two critical flaws in PHP (CVE-2026-109618 and CVE-2026-109619) could let an attacker run code on a vulnerable server without logging in. PHP powers roughly 70% of websites that use a known server-side language, and fixes shipped on October 21 (Fox News). Details leaked a day before the fix. Government portals in the Netherlands, Finland and Belgium went offline as a precaution (ScruteX). By October 26, CISA had added both flaws to its list of vulnerabilities being actively exploited.

The twist? The flaws were found by OpenAI’s agents (Axios). AI found the crack. Then the whole internet, attackers included, sprinted to the same crack.

2. Cloud takeovers in under three hours

Microsoft reported that a China-aligned group, Storm-0940, used an open-source penetration-testing framework against hundreds of organisations. In several cases, attackers reached privileged cloud access in less than three hours. In one password-spraying campaign, Microsoft logged over 151,000 failed logins, peaking at about 2,900 attempts per minute (Fox News).

That’s not a person guessing passwords. That’s a machine that never takes a chai break.

3. The AI boom’s plumbing got poisoned

A malicious version of LiteLLM, a popular open-source gateway that connects apps to AI models, was published using stolen GitHub tokens. It harvested cloud, database and AI keys. The package averages about 3.3 million downloads a day (ScruteX).

Developer hosting platforms got hit too. Railway confirmed attackers accessed and partly decrypted customer environment variables. Zeabur exposed 321,810 accounts and 3.2 million environment variables. OpenCode disclosed that AI provider keys for about 1.4% of its Console users may have been exposed (ScruteX). That last one is where our $449 founder comes in.

4. WordPress, the marketer’s favourite playground

Malware showed up in Rank Math PRO plugin downloads on October 15–16. Separately, stolen connection keys for UpdraftCentral were used to create unauthorised accounts and plant malicious plugins on connected websites (Secucenter). If those names sound familiar, it’s because they sit on a lot of marketing websites.

5. Closer to home: India got hit too

A dark-web dump claimed to hold about 3.75 lakh State Bank of India records. SBI’s stock exchange filing on October 21 confirmed 3,897 customer records with personal details. Indian certificate authority eMudhra was also found to have generated intermediate certificate keys from weak values, so around 30,000 emSign certificates now need replacing (ScruteX).

Add a Lazarus-linked espionage campaign against South Korean organisations (Secucenter) and a joint FBI–CISA warning about SafePay ransomware, with at least 600 victims since August 2024 (ScruteX), and you get the picture.

The plot twist: the sky didn’t fall. This time.

Here’s the part most viral posts skipped.

Millions of hacking attempts produced surprisingly few successful attacks. Axios reported that the PHP flaws proved unusually hard to exploit, and some researchers compared the week to Y2K, when loud warnings pushed companies to fix things in time (Axios).

ScruteX also found that early coverage got details wrong in 12 of the 28 incidents. The European government shutdowns were precautionary, not attack damage. And some clips claimed SBI had lost all its customers’ accounts and passwords (Yahoo Finance), but only 3,897 records are confirmed (ScruteX).

So was it all hype? Not quite. Think of it as a fire drill where the fire was real but the building was mostly fireproof. Mostly.

And here’s the uncomfortable lesson: panic is also an attack surface. Every scary headline is raw material for a scammer. ScruteX expects SBI-themed scam calls and fake loan offers built on the leaked details (ScruteX).

Why this time feels different: AI is playing both sides of the chessboard

Cyberattacks aren’t new. What changed in October is the speed.

Security used to run on a rough rhythm. Someone finds a bug, a patch comes out, companies get a few weeks to update, and attackers slowly build working exploits. AI just squeezed that rhythm into days, sometimes hours.

Three shifts stand out:

Finding bugs is now automated. AI agents are finding critical flaws in software that has been around for decades (Axios). Great for defenders, if defenders get there first.
Attacks run at machine speed. Thousands of login attempts a minute. Cloud takeover in under three hours (Fox News). An incident response plan that starts with “let’s schedule a call” is already late.
API keys are the new crown jewels. Look at what attackers went after: AI keys, cloud secrets, environment variables, developer tokens (Lumu). In the AI era, a leaked key isn’t just a data risk. It’s a credit card with no limit and your company’s name on it.

The good news? The same AI that finds the holes is helping patch them. The PHP Foundation used AI-assisted tools from Anthropic and OpenAI during its own research (Fox News). This is an arms race, not a lost war.

“I’m a marketer. Why should I care?”

Because you probably own more of the attack surface than your CISO realises.

Think about what a typical marketing team runs today:

A WordPress website on PHP, with SEO, backup and form plugins installed by three different agencies over five years.
AI tools everywhere: content generators, chatbots, voice agents, enrichment tools. Each one connected with an API key someone pasted into a settings page in 2024.
Automations and integrations linking your CRM, ad accounts, email tool and spreadsheets.
Vibe-coded landing pages and microsites spun up on hosting platforms in an afternoon.
Customer data, lots of it, flowing through every one of those tools.

Every one of those is a door. Black October showed that attackers are rattling every door, automatically, all the time.

There’s a positioning lesson here too. When a bank’s customers start getting scam calls built on leaked details, nobody blames “the vendor”. They blame the brand. In the AI era, trust is a feature, and security is part of your go-to-market. The companies that communicate clearly and calmly during a scare will win customers from the ones that go silent.

The Black October checklist (do it before November eats your week)

No jargon. Just the stuff that matters.

Patch PHP. Ask your developer or hosting provider to confirm you’re on 8.2.35, 8.3.36, 8.4.27, 8.5.12 or later (ScruteX). If your site runs an end-of-life version, that’s a conversation for this week, not this quarter.
Audit your WordPress plugins. If you use Rank Math PRO or UpdraftCentral, look for unknown admin accounts and unfamiliar plugins (Secucenter). Delete anything you don’t actively use.
Rotate your secrets at the source. If a key lived on an affected platform, revoke it with the provider that issued it (OpenAI, AWS, your CRM), not just where it was stored.
Put spending caps and billing alerts on every AI key. That $449 surprise could easily have been $4,490. A two-minute setting saves a very awkward finance conversation.
Switch to passkeys or MFA everywhere. Password spraying only works on accounts protected by passwords alone (Fox News).
Inventory your connected apps. That integration from a 2023 hackathon still has access to your CRM. Kill it. (If your AI stack has sprawled, here’s how to audit your marketing stack.)
Prepare your team for the scam wave. Any “urgent security” message asking for a click, an OTP or a payment is guilty until proven innocent.
Know your reporting clock (India). CERT-In expects specified cyber incidents to be reported within 6 hours of noticing them (ScruteX). Someone in your team should know who to call before anything happens.

The bottom line

Black October wasn’t the apocalypse. It was the trailer.

The real story isn’t that hackers had a good month. It’s that AI has permanently changed the speed of the game. Bugs get found faster, attacks run faster, and the gap between “we’ll fix it next sprint” and “we’re in the news” keeps shrinking.

The companies that come out ahead won’t be the ones with the biggest security budgets. They’ll be the ones where everyone, marketing included, treats security as part of the job.

So go check your plugins. Rotate that key. Turn on the billing alert.

November is watching.

PS: If your first instinct was “let me forward this to IT”, fair enough. But first, open your AI tool settings and count how many API keys you personally created this year. Then forward it.

PPS: October is also Cybersecurity Awareness Month. You genuinely cannot write irony this good.