AI-built attacks hit 105 websites, with one domain taken for $12

Attackers are using AI models to generate large volumes of attacks for very little money, according to a report published by Dark Reading on 23 October 2026. The report describes one campaign that targeted 105 websites using open-weight models, and a separate case in which publicly available AI models were used to gain full control of a domain for $12.

Open-weight models are AI systems whose parameters are published, so anyone can download and run them without going through a provider's service. That matters for defenders because such models can be run locally, outside the safety filters and usage monitoring that hosted services apply.

The second case cited by Dark Reading is the more striking of the two on cost. Public models — those offered as a service — were used to obtain full domain control, meaning administrative authority over a target's domain, for a total spend of $12. The report does not name the targets of either campaign, and no vendor or product has been identified as the source of the weakness.

Dark Reading frames the two cases as evidence that cyber-defence will have to change, because the economics of attacking have shifted. Where a campaign against 105 separate sites would once have required substantial human effort, the report indicates that model-driven tooling can now produce that volume of activity cheaply.

Details remain thin. The incident record compiled from the reporting lists the root cause as tool misuse — attackers using AI tools as intended, but for hostile ends, rather than exploiting a flaw in the models themselves. It does not record a date for when the attacks took place, only that they were reported on 23 October 2026. The severity is assessed as moderate. The status of the campaigns is listed simply as reported, with no confirmation of remediation.

For security teams, the practical question raised by the report is one of volume rather than sophistication. Nothing in the account suggests a novel attack technique. The claim is that familiar techniques can now be generated and directed at many targets at a cost low enough that the usual assumptions about attacker effort no longer hold. The Agentic Times has not independently verified the figures, and further detail on how the $12 domain takeover was achieved has not been published in the material reviewed.