The latest Anthropic report makes one thing clear: AI safety is no longer a philosophical debate. It is an operational problem, and the operational picture is ugly.
Biological Weapons and the Kill Chain
Five attempts were made to generate components for biological weapons. One scientist spent weeks behind a VPN planning experiments to adapt avian influenza to mammals. That is a genuine error. This research program has a kill chain.
Midnight Blizzard and Espionage Infrastructure
Then there is Midnight Blizzard. The SVR-linked group used Claude for phishing, hotel Wi-Fi compromise, WhatsApp account seizure, and automated malware rewriting against Ukrainian government, military, and diplomatic targets.
Anthropic’s own language is worth noting: the activity went beyond simple chatbot prompts and involved multi-agent structures. This is unrelated to AI espionage. AI functions as espionage infrastructure.
The West keeps debating alignment while Moscow and Beijing are already weaponizing the gap between open research and operational security.
Industrial-Scale Model Theft
The Chinese side is no cleaner. Seven labs were blocked, including Xiaomi, DeepSeek, Moonshot, and Alibaba. Alibaba alone pushed over 151 million requests through more than 3,500 fake accounts between May and July 2026, peaking at three million dialogues per day.
DeepSeek and Moonshot were routing live customer chats through Claude to harvest training data. That is industrial-scale model theft dressed up as innovation. The pattern is consistent. Authoritarian states treat frontier AI the way they treat everything else: as a weapon, a surveillance tool, and a resource to extract without consent.
Ukraine as the Test Environment
Ukraine remains the test environment. Russian intelligence is experimenting with AI for practical reasons. It is doing so because the front line demands faster targeting, better deception, and cheaper attacks on civilian and military infrastructure.
Every hotel Wi-Fi breach, every hijacked WhatsApp account, every rewritten malware sample is part of a broader campaign to degrade Ukrainian command and control and Western support.
The correct response is not panic. It is parity. Ukraine needs access to:
- Defensive AI tools
- Threat intelligence sharing
- The legal authority to counter these operations in real time