Claude was used in cyberattacks and scams, Anthropic reports

Attackers handed over more of the work

A suspected Russian espionage operator used AI to build attack tools, run phishing operations and organize stolen information, according to Anthropic’s September report. Claude was involved throughout the campaign.[1]

Humans chose targets and reviewed stolen data, while AI handled much of the work in between.[1]

The attack reached hotel guests

One route to victims ran through hotel Wi-Fi. Microsoft’s July 31 investigation describes a campaign it calls CaptiveCrunch, in which attackers manipulated hospitality networks to redirect travelers to malicious sites. Some visitors encountered fake browser or operating-system updates that delivered software capable of stealing files, recording keystrokes and taking control of a device.[3]

Microsoft linked the campaign to Midnight Blizzard and reported extensive AI use. Anthropic’s account describes compromises of at least three hotel Wi-Fi providers.[3][1]

The familiar act of connecting to a hotel network became an opportunity to reach a person’s device. The attack depended on making a malicious instruction look like part of a service the traveler was already trying to use.

Software that could adapt during a campaign

Anthropic also describes AI checking whether security products detected the operator’s malicious software, then modifying and rebuilding it to evade those defenses.[1]

This is the significant change in the account: blocking one version of an attacker’s software may buy less time if producing the next version requires less human work. The pressure falls on defenders to recognize the broader activity as well as the individual program.

Anthropic selected notable operations disrupted between December 2025 and August 2026, including scams and surveillance. The report does not measure how common misuse is.[1]