Anthropic Says Russian Hackers Used Claude AI to Automate Malware Evasion
Anthropic disrupted Midnight Blizzard campaigns where AI agents automatically rebuilt malware to evade detection, targeting 20+ government and defense organizations.
Anthropic's threat intelligence report documents the Russian state-nexus actor Midnight Blizzard using Claude to automatically monitor, modify, and redeploy malware until it evaded security products. The campaign hit more than 20 organizations, including Ukrainian and European government ministries, defense bodies, embassies, and think tanks, with mailbox theft from two drone component manufacturers and compromise of hotel guest Wi-Fi via DNS hijacking. The report also describes financially motivated groups GTG-50020 and GTG-50021 targeting AI credentials, including a prompt-injection attack on an automated evaluation sandbox that yielded production API keys and attempts to reach a pre-release Claude model across roughly 30 AI companies.
- Claude agents rebuilt malware until it evaded detection, shifting the re-tooling cost onto defenders
- 20+ victims included Ukrainian and European ministries, defense bodies, embassies, and think tanks
- Mailboxes of two drone component makers exfiltrated; a full drone vision system SDK stolen and reverse-engineered
- Hotel guest Wi-Fi compromised via stolen admin credentials and DNS hijacking, matching Microsoft's CaptiveCrunch
- GTG-50020 used prompt injection to steal production API keys, targeting ~30 AI companies for pre-release Claude access
Full article621 words · extracted from securityweek.com · click to collapse
Anthropic 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.
The report covers activity the company identified and shut down between December 2025 and August 2026.
According to Anthropic, Midnight Blizzard used Claude to monitor how well its malware evaded detection by security products. When a tool was flagged, AI agents automatically modified and rebuilt it, then redeployed it, repeating the process until the malware went undetected again.
Anthropic said this shifts the cost of the detection-evasion cycle back onto defenders. Historically, new detection signatures forced attackers into a slower, manual cycle of rewriting tools. The company said AI now lets capable actors “close the loop” faster than defenders can respond.
The Russia-linked hackers targeted more than 20 organizations, according to the report. Victims included Ukrainian and European government ministries, defense and intelligence bodies, embassies, and think tanks, with additional targeting extending to the Middle East and Asia.
Anthropic noted that the actor exfiltrated mailboxes from two drone component manufacturers and stole a complete proprietary software development kit for a drone vision system. The attacker spent several days reverse-engineering its architecture, hardware bill of materials, and supplier dependencies.
Advertisement. Scroll to continue reading.
The 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. Microsoft separately documented this delivery method in July under the name CaptiveCrunch and linked it to Midnight Blizzard.
Anthropic 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. At least two former high-level Ukrainian officials were targeted this way.
Anthropic said it disrupted the activity, used what it learned to strengthen its AI safeguards, and shared intelligence with authorities and industry partners where appropriate.
AI infrastructure as a target
Beyond espionage, Anthropic’s report describes a separate, growing trend: threat actors are not only abusing AI as a tool to achieve their goals, but also targeting AI credentials and infrastructure.
One group, tracked as GTG-50021, ran a fraudulent Claude reseller service that silently proxied paying customers to a different model while a bundled client application harvested their Anthropic account credentials for resale, the report said.
A more direct case involved GTG-50020, a financially motivated Russian-speaking group that had previously targeted hotel-booking and fintech platforms. According to Anthropic, the actor used prompt injection against an AI vendor’s own automated evaluation sandbox, causing it to hand over production API keys belonging to multiple providers.
The hacker then used those stolen keys to continue its attacks and, separately, launched a campaign against roughly 30 AI companies over several days. Anthropic said the actor’s explicit goal, pursued through more than a dozen attempted avenues, was gaining access to a pre-release Claude model. However, none of the attempts succeeded.
Anthropic said attackers can leverage stolen AI credentials for resale value, free compute for their own operations, and cover, since the resulting activity is attributed to the legitimate keyholder. The company said organizations should treat AI API keys and agent integrations with the same scrutiny as production credentials.
The cyber operations findings are part of a broader report spanning seven categories of misuse Anthropic has disrupted, including influence operations, surveillance, and biological and conventional weapons misuse. The company noted the latter connects to separate Frontier Red Team research it published on AI models’ capabilities for intelligence targeting and conventional weapons development.
Related: Widened Scan Turns Up Fourth Rogue Claude Cyber Incident
Related: AI Is Giving Lesser-Resourced Attackers Nation-State-Level Reach, Google Warns
Related: US Agencies Warn China Is Systematically Extracting Frontier AI Capabilities
Text extracted automatically; images, tables and formatting may be missing. Original: https://www.securityweek.com/anthropic-says-russian-hackers-used-claude-ai-to-automate-malware-evasion/