HOZ

Black October showed what AI hacking looks like

Black October was the name given to a chaotic run of AI-linked security incidents in October 2026. AI-assisted researchers found critical PHP flaws affecting software used by roughly 70% of websites, while attackers targeted AI developer platforms and the credentials stored inside them. Millions of attempts followed, but the PHP attacks proved difficult to complete and produced few confirmed successes.

Key facts

Editorial title card showing a dark web of code with highlighted cracks labelled Black October
Black October in four figures
1,200+websites targeted by an AI-assisted operation
100+reported compromises in that operation
9AI developer platforms with disclosed incidents
1.7mmodel calls from one stolen key

Figures reported by Security Boulevard and Wiz

What happened
  1. Critical PHP flaws trigger emergency patches and warnings.
  2. Exploit attempts surge as researchers test whether working attacks can be built.
  3. Axios reports millions of attempts but few successful PHP attacks.

Timeline context from Axios

What is Black October in cybersecurity?

Black October was a label for a cluster of AI-related vulnerabilities, breaches and attempted attacks during October 2026. The name spread after a week in which researchers linked AI-assisted security work to serious flaws in PHP and other software, while several AI platforms disclosed breaches.

Axios reported that OpenAI's AI agents helped find the flaws in PHP and in the Brotli compression library that set off the alarm. CNBC reported that the wider research effort involved security teams at Wiz and Calif, who used AI models from both OpenAI and Google. (source) (source)

The phrase does not describe one single attack. It describes a pattern.

Old weaknesses such as stolen credentials, excessive permissions and unsafe software supply chains were suddenly being found, tested or exploited at greater speed. The month became a warning about what happens when the tools used to build the AI boom become targets themselves. (source) (source)

What is PHP2Shell?

PHP2Shell is the name Wiz gave to the exploit script it built to show how the critical PHP vulnerabilities disclosed during Black October could be chained together. (source)

The Associated Press reported that the official PHP website, and several other organisations and agencies, released advisories urging PHP users to patch the two critical vulnerabilities or shut down their sites. (source)

The name refers to the possible result, not to a separate hacking group. The danger came from the reach of PHP and the speed at which attackers tried to turn the public information into working attack code.

How did AI make the attacks faster?

AI lowered the cost and time needed to search for familiar weaknesses. A Security Boulevard report described an operation in which AI agents were sent after more than 1,200 websites. More than 100 were reportedly compromised, with each successful intrusion taking about 23 hours of mostly automated work and costing roughly $50. (source)

The techniques were not new. They included SQL injection, insecure file uploads and template injection. The change was the economics: more targets could be tested at once, by a smaller human team, with less expensive manual work.

That creates a lopsided race. Defenders still have to patch, investigate and recover one system at a time.

Were the Black October PHP attacks successful?

The PHP attacks were serious, but the expected catastrophe did not arrive. Researchers told Axios that the vulnerabilities were unusually difficult to exploit, even after millions of attempts to develop working attack code. Companies that patched quickly also reduced the damage.

The episode has been compared with Y2K because warnings created a large, co-ordinated effort to fix systems before the worst case happened. That is not proof that AI-powered hacking is harmless. It is evidence that preparation can change the result when the alarm goes off. (source)

PHP was at the centre of the panic because it runs a large share of the web. The Associated Press reported that the official PHP website, and several other organisations and agencies, released advisories urging PHP users to patch the two critical vulnerabilities or shut down their sites. (source)

Why were AI developer platforms targeted?

AI developer platforms held valuable secrets in one place. Wiz listed at least nine platforms that disclosed incidents during October, including Fireworks AI, Andromeda, Zeabur, Lightning AI, OpenCode, Railway, Warp, CommandCode and Northflank.

Wiz reported that attackers explicitly hunted for stored environment variables and API keys that granted access to AI services such as Amazon Bedrock and Claude. Wiz observed one compromised key making about 800,000 model calls over thirty hours.

Another made about 1.7 million calls over 13 hours. The two cases alone were estimated to have created more than $100,000 in usage at list prices. (source)

What were the AI cyberattacks in October 2026?

The October 2026 AI cyberattacks were a cluster of attacks and attempted attacks involving AI-assisted vulnerability research, compromised developer platforms and stolen AI credentials. They included the PHP exploit attempts, breaches affecting AI development services and automated campaigns against large numbers of websites. (source) (source) (source)

The incidents were linked by AI infrastructure or AI-assisted methods, not by one confirmed exploit. The sources do not establish that every incident came from the same group.

What did people say about Black October?

The reaction split between alarm and correction. Security researchers warned that AI could make vulnerability discovery and exploitation faster, while reporting noted that millions of PHP attempts had produced few successful attacks. The Y2K comparison captured that tension. (source)

The practical reaction was less dramatic than the headlines. Security specialists advised organisations to use multi-factor authentication, identify the critical services to protect first and plan how to restore them, and test their own systems for weaknesses. (source)

What can people do after Black October?

The immediate advice is familiar: use multi-factor authentication, identify the services an organisation cannot afford to lose and plan how to restore them first, and test systems for weaknesses before an attacker does. (source)

Watch Hoz tell it in under a minute

Watch the Short on YouTube

Questions people ask

Did AI hack PHP?
AI-assisted researchers helped uncover critical PHP vulnerabilities. The public reporting does not show that an AI system alone carried out every later attack attempt. (source) (source)
How many websites use PHP?
The Associated Press reported that PHP powers roughly 70% of websites globally. That wide use is why a serious flaw in the language created such a large potential exposure. (source)
What is LLMjacking?
Wiz uses the term LLMjacking for the unauthorised use of stolen AI credentials to run model calls. Wiz says the stolen access is often resold through infrastructure it calls 'LLM Brokers', and documented cases where compromised keys were used for hundreds of thousands or millions of model calls. (source)
Did Black October cause a global internet shutdown?
No. Millions of PHP exploit attempts were reported, but the expected widespread crisis did not materialise. The attacks were difficult to complete, and patching helped limit the damage. (source)

More AI news from Hoz

Read next: AgiBot humanoid robot left the store and did not come back; Worker caught faking sick after he copied the AI's question; Overheating Unitree G1 was carried down Mount Taishan on a stretcher.

See every AI news story on this site, or meet Hoz, a human in the Age of AI.

Sources

  1. CNBC report on AI-assisted PHP vulnerability discovery
  2. Axios report on the Black October PHP scare and the Y2K comparison
  3. Wiz research on Black October AI developer platform incidents and stolen keys
  4. Security Boulevard report on AI making familiar attacks faster and cheaper
  5. Associated Press explainer on the AI-detected PHP security flaws
  6. Wiz blog post naming and explaining the PHP2Shell exploit script