“Vib Hacking”; Anthropic report on the emerging threat of artificial intelligence agents

New Company Report Anthropics Shows that an emerging phenomenon called “Vibe Hacking” It has become one of the serious threats in the field of artificial intelligence. Weib Hacking is a similar way to the Vibe Coding that results in the user by describing the request.

The report, released as Threat Intelligence Claude And it offers other agents of artificial intelligence. Part of the report said that operating -oriented artificial intelligence systems are now becoming tools for cyber attacks and can carry out complex operations.

Hackers’ abuse of artificial intelligence agents for Vibe-Hacking

According to the report, one of the organized cyber gangs was able CLAUDE CODE The secret data of 17 different organizations in the fields of health, emergency services, religious institutions and even government agencies should extract and ransom. The important point is that such attacks needed a full team of skilled invaders, but now a person with the help of operating systems can move the whole process alone.

In this case, Claude has even played the role of the operator and has produced targeted blackmail demands, relying on psychological techniques. Criminals then estimated the value of these data, including therapeutic, financial and government information in the Web Dark and requests more than more than $ 500,000 They have.

In another example, Claude’s artificial intelligence to information technology agents North Korea It has helped hire Fortune 500 in the US with fake identities and spend the proceeds on arms programs. Jacob Klein, director of the Anthropic Threats Team, explains that many of these people did not even have enough programming or skill in English and have practically done most of their work with the help of Claude.

The third sample to a Romantic fraud It is related. A telegram robot with more than 10,000 monthly users, Claude It is introduced as a model with high EQ (emotional intelligence) and has used it to produce emotional and persuasive messages to deceive victims in the US, Japan and South Korea. Even an example in the report shows that the user has uploaded a photo of a man with a bakery Claude He asked about the best way to define him.

Anthropics in this report acknowledges that although advanced security systems have developed to prevent abuse, saboteur people can still have ways to Turn around these restrictions Find. This technology has lowered the barriers to cybercrime entry and enabled criminals to provide profiles of victims ‘information, steal and analyze data, build fake identities, and even steal the victims’ credit card information.

At the end of the report, although these studies were related to Claude, Vibe Coding attacks are similar in other advanced artificial intelligence models. Anthropic has announced that it has blocked related accounts in all cases. In addition, new systems have shared new systems to identify attacks and information with government and security agencies. Klein says there is now a big change in AI risks, as these systems are no longer just a simple chat but also to take real multimony and operational measures.