Anthropic published its first public threat intelligence report Thursday, detailing how state-backed hackers, criminal groups, and hacktivists tried to misuse Claude between December 2025 and August 2026. The most striking case involves a cluster the company linked to Midnight Blizzard — also known as APT29, Cozy Bear, and BlueBravo — a hacking group Western intelligence agencies attribute to Russia’s Foreign Intelligence Service (SVR). After compromising hotel Wi-Fi providers and redirecting travelers to attacker-controlled infrastructure, the group focused squarely on Ukraine’s defense ecosystem. Hackers broke into the mailboxes of two drone-component manufacturers and eventually stole a complete proprietary software development kit for a drone vision system. They then fed that code into Claude to recover the product’s architecture, hardware bill of materials, supplier dependencies, and details of an unannounced product. “Military drone control and AI vision-related firmware appeared to be of particular interest,” the report noted.
The same group also used Claude as a real-time detection-evasion tool. Whenever a security product flagged one of their implants, the operators fed the flagged artifact into Claude, which modified and redeployed it — effectively closing the feedback loop between detection and re-evasion faster than defenders can push new signatures. Anthropic’s report concludes that “AI has inverted the cost back onto defenders”: a capability once measured in days of manual rewrite time is now compressed to minutes. Anthropic said it disrupted the activity in each case and shared indicators of compromise (IOCs) and threat intelligence with authorities and industry partners. The report also documents ShinyHunters affiliates using Claude to scan for credentials and map unfamiliar cloud environments — in one case moving from a stolen developer token to full administrative access in roughly three hours.