Tricks of the Hacking Trade
Never connect to a hotel’s Wi-Fi. That’s one lesson from a report Anthropic published Thursday on cyberhacks it had identified involving its technology.
The report—which makes a fetish of jargon, like “generative threat groups” or GTGs, for bad people—described one hack where, among other things, the bad guy hacked firms that operated hotel Wi-Fi systems for guests. People who then connected to the hotel Wi-Fit had details of their device (and traffic) sent to the bad guy’s servers, while malware went to the victims’ phones or computers. AI was used at “every point” in the operations.
That was one of the more pedestrian hacks described in the report, which delved into a variety of ways bad guys are using Claude to do everything from steal AI tokens to spread misinformation and influence opinion, particularly in countries such as Ukraine. Some of the examples are scary: The report highlights five examples of criminals using Claude “in ways that could support” development of biological weapons.
Anthropic took action to block the uses of its models for those purposes, to be sure, but these examples are perhaps one reason why folks connected with Anthropic have been warning this week about the dangers of the technology. There’s a lot in the report, including information about Chinese AI startups Moonshot and DeepSeek sending user requests to Claude for answers without telling users. It’s worth reading.