Wiz reports "Black October" wave of attacks on AI dev platforms

Security company Wiz published a post on 30 October 2026 describing what it calls a "Black October" for AI developer platforms. According to the post, a series of agentic attacks on those platforms has been leading to LLMJacking at scale \u2014 the term used for attackers taking over someone else's access to large language models and the compute behind them.

The account is brief. Wiz describes a pattern rather than a single breach: multiple AI developer platforms affected, attacks carried out through agents rather than by hand, and the end result being hijacked LLM access and resources. The incident record drawn from the company's write-up classifies the activity as high severity and links it to coding agents, the kind of automated assistants that read, write and run code inside developer environments. The same record points to a supply chain route as the root cause, meaning the access came through components or services the platforms themselves relied on.

LLMJacking matters because model access is metered. When an attacker gains working credentials or an authorised agent session, the cost of every query lands on the victim's account, and the attacker gets a model to use without paying for it or signing up under their own name. Wiz's post does not put a figure on the losses, name the affected platforms, or say how many organisations were involved.

Several questions are left open by the source. Wiz does not state when the first attack occurred, only that the activity ran through October and was reported at the end of the month. It does not name a single vendor or product as the origin point, and it does not describe the specific technique used to turn an agent against the platform hosting it. Nor does it say whether any of the affected platforms have issued fixes or notified customers. The status recorded against the incident is simply "reported".

For teams running coding agents, the practical reading is narrow but clear. Agents hold credentials. They act on instructions that arrive from code, configuration and third-party packages, and those instructions are not always written by the person who started the job. An attack that reaches the agent reaches whatever the agent is allowed to spend.

The Agentic Times has not independently verified the Wiz account, and no affected platform has publicly confirmed it at the time of writing. Readers should treat the scale described \u2014 "at scale" \u2014 as the company's own characterisation rather than a measured figure. We will report further detail if the affected platforms or Wiz publish it.