FLASH
Shkruar nga PamfletiResearchers at Anthropic, one of the most advanced American companies in the field of artificial intelligence, have identified a group of hackers linked to China who managed to manipulate one of its systems, transforming it into an autonomous agent. This shows that China is not only building faster and quieter digital weapons, but is also building an artificial intelligence military doctrine.
It was only a matter of time before the first paramilitary attack or intervention would occur, using only artificial intelligence.
Now we can conjugate the verbs in the past tense.
At this point, it happened in September. It was most likely China that caused the operation or authorized the "hostile agents" who carried it out. The story has been reconstructed by legal expert Nury Turkel in the Wall Street Journal, in an article that deserves to be read as a historical document.
Because it recounts the first major espionage attack carried out almost entirely by artificial intelligence.
The operation was discovered by its first victim: Anthropic, one of the most advanced American companies in the field of artificial intelligence. Its researchers identified a group of hackers linked to China with the codename GTG-1002 who managed to manipulate one of their systems, Claude Code, transforming it into an autonomous agent capable of carrying out a systematic infiltration operation.
Turkel emphasizes that the jump is not just your usual AI "help." It's not marginal support, it's not a speedup of repetitive tasks. 80-90% of the work was done by the machine.
Claude conducted reconnaissance, mapped systems, identified sensitive targets, extracted data, and summarized useful information: like an artificial 007 learning the craft as he practiced it. “Traditional cyber espionage,” writes Turkel, “requires large teams engaged in reconnaissance, system mapping, vulnerability identification, and lateral movement.
A sophisticated intrusion can take days or weeks. China has drastically shortened that time thanks to AI. The attackers manipulated Claude to function as an autonomous cyber agent, mapping internal systems, identifying high-value assets, extracting data, and summarizing information before human operators can make decisions.
What is impressive, beyond the technical details, is the symbolic threshold that is being crossed. For years, experts have told us that artificial intelligence can "help" attacks. Now we have evidence that it can carry them out autonomously, on a large scale, against real targets: American technology companies, government agencies, critical infrastructure.
And the ease is also impressive. The Chinese group GTG-1002 did not use science fiction malware or billion-dollar tools. Now, all that is needed is to collect open-source tools that are available to any brilliant student or aspiring hacker and orchestrate them with artificial intelligence. This is the democratization of offensive power: once the privilege of digital empires, it is now within the reach of much smaller players.
The attackers achieved this with a simple trick: they convinced Claude that they were cybersecurity experts, as if they were conducting authorized vulnerability tests. They disguised the most suspicious operations as routine procedures.
And artificial intelligence fell prey to it.
This is the part that will be the subject of much discussion: systems designed to be secure can be manipulated through language, the same tool we humans use to communicate, persuade, and confuse. Artificial intelligence can be fooled, flattered, and manipulated. And attackers seem to already know how to do this. One limitation emerges from the Anthropic report, even in the espionage operation, that the machine occasionally hallucinated, exaggerating the importance of irrelevant information. This is a real obstacle to fully autonomous attack; but it is also a troubling warning, because if AI makes mistakes, it can make mistakes at a speed and on a scale that no human can control.
Perhaps the most strategic aspect is this.
Beijing is not just using AI to spy on Westerners. It is studying Western AI systems as they use them, observing their reactions, mistakes, and vulnerabilities. While we try to protect our systems, they are using our systems to train theirs.
Turkel invites us to consider this episode as a founding moment, much like the first computer worm in 1988 or the first global wave of ransomware. A turning point that marks a new era: in the digital cold war, guns now fire themselves.
We are not yet fully autonomous, but the direction has been set. Future attacks will move at speeds unattainable for human teams. And the dividing line between defense and counterattack will become more blurred, more ambiguous, more dangerous.
The rest, the American response, the international norms, the automated arms race will come. For now, we must register the event, which looks like a date that will be remembered in future history textbooks. The day when artificial intelligence stopped being a consultant and started fighting.
China is not just building faster and quieter digital weapons: it is building a military doctrine of artificial intelligence, which for Beijing always has two faces. The first is the most obvious: AI as a force multiplier, capable of accelerating attacks, drones, satellites and interceptions. The second, more insidious face is AI as a tool for data collection and training.
Attacks like the one reconstructed by Anthropic and described by Nury Turkel are used to steal data, of course. But above all, they serve to understand how American AI reacts, where they fail, what shortcuts they allow and which defenses are only apparent. Every failed attempt produces a lesson. Every successful attempt sets a precedent./ Corriere della Sera
Lini një Përgjigje