Artificial intelligence is no longer limited to generating text, images, or code. With Claude Mythos Preview, Anthropic has shown the world a model capable of identifying critical vulnerabilities in computer systems and autonomously creating exploits. A technological step that opens up enormous possibilities, but also raises increasingly urgent questions about the future of digital security.
According to Anthropic, Claude Mythos was able to identify thousands of vulnerabilities in browsers and operating systems without direct human intervention. This is not simply an advanced chatbot, but a system designed to reason in a sophisticated way within complex environments, understanding infrastructures, code, and potential weak points.
The real novelty, however, is not only technical.
For the first time, we are seeing a concrete glimpse of a new generation of AI capable of automating activities typically associated with offensive cybersecurity. And it is precisely this aspect that has captured the attention of researchers, companies, and governments.
AI is no longer just assisting humans: it is starting to make operational decisions.
A change that could completely redefine the relationship between cyber attack and defense.
The first tests
Anthropic has chosen to limit access to Claude Mythos to a small number of selected partners, openly stating that it wants to prevent the misuse of the technology. This decision clearly reflects the level of sensitivity the sector has reached. If similar tools became accessible without controls, the risk would be a massive acceleration in the offensive capabilities of criminal groups and cyber attackers.
In recent months, AI safety has become a central topic in the international technological debate. Not only because of the potential of language models, but also because these technologies are acquiring increasingly sophisticated operational capabilities. From the automated generation of phishing attacks to the autonomous discovery of vulnerabilities, the line between intelligent assistance and offensive automation is rapidly becoming thinner.
And this is where the real critical issue emerges: traditional cybersecurity may no longer be enough.
A new era of cybersecurity
Companies must no longer defend themselves only against human attacks, but also against intelligent systems capable of learning, adapting, and acting at speeds impossible to replicate manually. This means rethinking infrastructures, processes, and protection strategies from an entirely new perspective.
Artificial intelligence, in fact, is not neutral.
Every innovation that enhances defensive capabilities can also potentially strengthen offensive ones. It is a dynamic the technology sector knows well, but today it takes on a much deeper dimension because it involves systems capable of evolving autonomously.
Per questo motivo la vera sfida non sarà soltanto sviluppare AI più potenti, ma costruire ecosistemi digitali in grado di governarle in modo sicuro, trasparente e sostenibile.
Il futuro della cybersecurity non dipenderà solo dalla tecnologia.
It will depend on the ability of organizations to understand how quickly the paradigm is changing.
Questo contenuto è stato realizzato nel rispetto dei principi di trasparenza e tracciabilità previsti dal Regolamento Europeo AI Act (2025). Tipo di contenuto: AI-assisted
