Anthropic Says Russian-Linked Hackers Used Claude to Automate Malware Evasion
Summary
Anthropic says a Midnight Blizzard-linked operation used Claude to iteratively modify malware to evade detection, while other actors targeted AI credentials and infrastructure. The company says it disrupted the activity and shared intelligence.
Key points
- Anthropic says it identified and disrupted activity attributed to the Russian state-nexus group Midnight Blizzard between December 2025 and August 2026.
- The group reportedly used Claude to check malware against security products and automatically modify and rebuild tools when they were detected.
- More than 20 organizations were targeted, including government, defense, intelligence, diplomatic and research bodies.
- Anthropic says the actor exfiltrated mailboxes from two drone component manufacturers and stole a proprietary drone-vision software development kit.
- The group reportedly used stolen hotel Wi-Fi administrator credentials for DNS hijacking and linked victims' WhatsApp accounts through headless browsers to export conversations.
- Anthropic says GTG-50020 used prompt injection against an AI vendor's evaluation sandbox to obtain production API keys; attempts to access a pre-release Claude model failed.
- Anthropic says it strengthened AI safeguards and shared intelligence, and advises treating AI API keys and agent integrations like production credentials.
Article Details
- Event Type
- Anthropic-reported cyberespionage and AI-enabled detection-evasion operation, alongside separate campaigns targeting AI credentials and infrastructure.
- Impact
- Anthropic said Midnight Blizzard targeted more than 20 organizations and exfiltrated mailboxes from two drone component manufacturers, as well as a proprietary drone-vision software development kit. The group also compromised at least three hospitality vendors to redirect guest traffic and took over victims’ WhatsApp accounts to export conversations. Separately, GTG-50021 harvested Anthropic account credentials, while GTG-50020 stole production API keys and attempted, unsuccessfully, to access a pre-release Claude model.
Threat Actors
GTG-50020Financially motivated Russian-speaking group that Anthropic said stole production API keys and attempted to access a pre-release Claude model.GTG-50021Group that Anthropic said ran a fraudulent Claude reseller service and harvested Anthropic account credentials for resale.Midnight BlizzardRussian state-nexus group that Anthropic said conducted the cyberespionage operation; Microsoft separately linked it to the CaptiveCrunch delivery method.
Vendors
AnthropicAnthropic disrupted a cyberespionage operation whose tradecraft and targeting match the Russian state-nexus group tracked as Midnight Blizzard, the company said in a threat intelligence report published this week.MicrosoftMicrosoft separately documented this delivery method in July under the name CaptiveCrunch and linked it to Midnight Blizzard.
Products
ClaudeAccording to Anthropic, Midnight Blizzard used Claude to monitor how well its malware evaded detection by security products.WhatsAppAnthropic said the same actor took over victims’ WhatsApp accounts by linking them as companion devices through headless browsers, suppressing read receipts to export conversations undetected.
Countries
Industries
DefenseVictims included Ukrainian and European government ministries, defense and intelligence bodies, embassies, and think tanks, with additional targeting extending to the Middle East and Asia.GovernmentVictims included Ukrainian and European government ministries, defense and intelligence bodies, embassies, and think tanks, with additional targeting extending to the Middle East and Asia.HospitalityThe group also compromised at least three hospitality vendors that operate hotel guest Wi-Fi, using stolen admin credentials to redirect guest traffic through DNS hijacking.intelligenceAnthropic disrupted a cyberespionage operation whose tradecraft and targeting match the Russian state-nexus group tracked as Midnight Blizzard, the company said in a threat intelligence report published this week.