A small group of Russian freelance developers have used the Claude chatbot to create software for an autonomous flock of impact FPV drones. It is able to independently select targets, coordinate the actions of several devices and carry out the final aiming. This is stated in the 154-page report of the Anthropic company developed by Claude, The Guardian writes. According to the report, the UAVs were programmed to independently select a target and crash into it, detonating without the final decision of a human. During the development of the system, the group repeatedly identified the same area of ​​the Donetsk region of Ukraine as a target. To bypass Anthropic's geo-restrictions, the developers used a VPN. Anthropic also reported an increase in the use of artificial intelligence by cybercriminals and state-linked hacking groups. According to the company, AB is used to automate a significant part of cyber attacks (including attacks against Ukrainian state bodies, military and diplomats). One of the groups was used by AB in almost all stages, from creating phishing attacks and capturing traffic in hotel networks to seizing WhatsApp accounts. Its methods matched those of the Midnight Blizzard hacking group, which the US government previously linked to Russia's foreign intelligence service. Hackers also created a system that automatically captured the detection of malicious code by security measures and overwrote it until it was no longer recognized. Next week, the US House of Representatives is expected to discuss a new package of sanctions against Russia. Its passage remains uncertain despite broad support in the Senate.