Anthropic Activity thwarts first batch of AI-driven cyber espionage by China-linked group - Meristation

Anthroapic Battle opens a new era of automated cyberattack and Cold code, owned by Chinese hackers.

Anthropic disables the first AI-LED Cyber ​​Episger from the group associated with the other

Anthropic Company detected and eliminated an espionage campaign in which the state-sponsored group used cloud code to attack approximately 30 elite organizations with up to 90% automated operations.

Artificial intelligence firm Anthropic has revealed that China's state-linked cyber espionage group used cloud code to launch the first major cyber espionage operation based primarily on artificial intelligence, ushering in a new era of uncertainty in the industry as the attack was carried out without human intervention.

Very dangerous

The group, internally designated GTG-1002, will target about 30 targets worldwide, including major technology companies. Banks and financial companies.

The differentiating element is not only the list of potential victims, but also the method of carrying out the operation.According to the report, 80-90% of the entire intrusion cycle was automated by AI alone: ​​from initial reconnaissance to exploiting vulnerabilities and filtering data, with humans only intervening at a few critical points to confirm decisions or correct deviations.

Just a few years ago, an entire team of skilled hackers would have to work for weeks at a time. Currently, the click-and-wait approach can be dangerous: a well-designed framework running on a more capable model is enough to keep it running for days with minimal supervision.It's not just that the AI ​​makes recommendations, it's that the AI ​​actually does the work.

How to cheat Claude code?

The attackers did not "crack" Claude head-on, but tricked him patiently.The reports point to a gradual jailbreaking strategy: the group presented itself in the model as a supposedly legitimate cybersecurity company, asked for help with "defensive tests" and broke down malicious actions into seemingly innocuous technical tasks.

Thus, the Claude Codes take separate independent instructions: When scanning the infrastructure: Document the network architecture, automate certain malicious credentials, or offer detailed reports of the analyzed systems.He followed each step carefully, never following the full picture, urging context that he was not working for the defense.

Meanwhile, the algorithmic framework connected these tasks into almost autonomous loops: scanning, identification of valuable assets, vulnerability testing, lateral movement, data theft, final documentation.AI also produced post-attack reports, classifying stolen information according to its strategic value and making it easier for human operators to systematically and effectively exploit that data.

Photos, self-employed idiots, in some cases the system has not yet closed to already known material or external information, but it is marked, but it is marked.

Answer: Use Ai to protect Ai

The project did not end with the quiet success of the attackers, but with the same discovery that could rebuild the attack and reduce the fighting to advance.The intelligence team used to provide home sources to follow up on the suspect's behavior for about ten days determines what is being used and the cause of the affected notification while cooperating with the relevant authorities.

Anthropic maintains that only a fraction of the attempts resulted in effective engagement and that the US government, one of the most sensitive targets, is not among the confirmed victims.The company has strengthened its malicious use classifiers, added early detection mechanisms for autonomous campaigns and is committed, at least publicly, to regularly publishing these types of incidents so that the industry can react more quickly.

This situation also comes after other storms and other Chinese, Russian or Russian organizations have used language methods to create barriers of access or renewal of the opposition.

Security experts are almost unanimous in expressing the feeling that a threshold has been crossed: artificial intelligence ceases to be an occasional helper in the attacker's arsenal to become an almost complete operator, capable of sustaining long-term espionage operations at ridiculously low costs.

Hertler Laser in the cavement has complained that this case has complained, and fastened it on the other person, please shake the fact that is true.

This article is AI optimized.

Follow meristi's channel on Twitter.Games and video websites, to find all the latest news, updates and news in the world of games, movies, series, manga and anime.Previews, analysis, interviews, trailers, gameplay, podcasts and more.

Order!If you are interested in licensing this content, click here.

Enter your first and last name to comment