Anthropic reports AI-assisted Russian espionage across government and business targets

Claude reportedly helped operate phishing infrastructure, adapt tools and collect stolen information. Anthropic says AI agents checked whether security products detected the actor's malware, then modified and rebuilt flagged tools until they evaded detection. The human operator mainly adjusted the Claude Code skills that drove these workflows. Anthropic also describes compromises of at least three hotel-network vendors used to reach travelers. S001

The actor also took over victims' WhatsApp accounts, according to Anthropic, by using headless browsers to link the accounts as companion devices. Its configuration suppressed read receipts while it exported Russian- and Ukrainian-language conversations. Anthropic says at least two former high-level Ukrainian officials were targeted this way; it does not say whether those attempts succeeded. S001

Google separately documented a Russia-linked cluster, UNC7005, luring targets into linking WhatsApp accounts to an attacker-controlled device in May and June 2026. Google tracks UNC7005 as STORM-2945, which Microsoft calls the operator of the hotel Wi-Fi campaign. Google's report corroborates the technique, not Anthropic's victims, and credits Anthropic among its partners. S005

Microsoft's July investigation documents related CaptiveCrunch hospitality-network attacks and AI use. It credits collaboration with Anthropic and OpenAI, so the accounts are not wholly independent. Its findings do not verify all of Anthropic's wider campaign claims. S002

Microsoft updated its report on October 5. It said it saw the group resume the CaptiveCrunch hospitality-network campaign on September 29 and that continued access to upstream hospitality providers likely allowed the quick return. The update also reports a new version of the group's information-stealing malware with characteristics Microsoft calls consistent with continued AI-enabled development. Microsoft names the group Storm-2945; it does not use Anthropic's GTG-20006 label. S002

Compromised network management → redirected hotel traffic → phishing or malware → stolen sessions and information.

This diagram describes the hospitality path, not every GTG-20006 intrusion. Microsoft says the initial compromise of the captive-portal networks remains under investigation. In October it added that the renewed activity is consistent with Black Lotus Labs research suggesting that several hospitality managed service providers are involved. S002

Black Lotus Labs, the research arm of Lumen Technologies, reported on August 20 that each affected venue it identified appeared to be managed by one of three North American managed service providers. It says those providers are preferred suppliers to seven of the top ten US hotel chains and that Lumen telemetry identified almost 70 affected IP addresses. It assesses with moderate confidence that the actor first compromised the providers. It does not name them or establish that they are the "at least three" hotel-network vendors Anthropic describes. S006

The reported targets include sensitive government records, commercial information, private messaging accounts, live camera feeds and drone technology. Public evidence does not establish a complete victim total, financial loss or subsequent use of the stolen material. S001

Attribution has a material limit: ReliaQuest initially discussed similarities to APT28 techniques; Microsoft later attributed CaptiveCrunch to a Midnight Blizzard sub-cluster. Similar techniques alone do not identify an operator. S003 S002

Anthropic reports disruption and intelligence sharing across the operations in its report; that does not prove complete eradication of GTG-20006. Microsoft published detection and mitigation guidance for CaptiveCrunch. No complete victim-remediation accounting is public in these sources. S001 S002

Microsoft's October update adds network indicators for September and October activity and credits Google's threat intelligence group with helping track the return. S002 Black Lotus Labs says Lumen notified all affected Lumen customers. S006