Latest

Sociological Analysis

Claude Mythos AI certifies the end of Modernity: we enter cyber-translimitation

Sociologist Javier Callejo explains why Claude Mythos AI means the end of Modernity and the forced entry into cyber-translimitation

Claude Mythos AI certifies the end of Modernity
Claude Mythos AI certifies the end of Modernity

We must begin by acknowledging that the concept of Modernity is complex and full of nuances. Even so, there are two aspects inherent to Modernity that most people could accept: its ambition for control and its confidence in human action. Well, both are radically transformed by the Claude Mythos Preview artificial intelligence model. The staging of the proposal itself has been revolutionary in many respects. A mise-en-scène that can be taken as a spectacle of the new technological power. An effective way of drawing the attention of the entire planet.

Claude Mythos and the end of Modernity

However, it is the announcement that certifies the end of Modernity, insofar as it is traversed by human control and confidence in humans. It is the announcement of the entry into cyber-translimitation. That limits have been crossed. In the piece by Observatorio Blockchain in which the news about this AI model was reported, it was made clear: the digital ecosystem has crossed the point of no return.

Let us turn to the procedure of the non-launch of the Claude Mythos Preview model. Anthropic, the company that created this AI model, decided not to release the project publicly for reasons of capability and risk. Until now, models that were not presented to society with great fanfare were not presented because they lacked final polish, still being in a prototype phase or similar.

Mythos, Anthropic's AI that has raised alarms over its destructive speed

Here Anthropic stated that Mythos had already been operating at full capacity. In this operation, and these were the alarms that accompanied the announcement, it was said that it had discovered thousands of serious vulnerabilities, including flaws in operating systems, browsers and open source projects. Flaws that have gone undetected for decades.

Control passes to machines

Mythos did not even have to break a sweat. It found them in seconds. It can find them in milliseconds and, most importantly, it requires intelligent machines that can repair them in those milliseconds. You cannot wait for the technical expert to get to work and take hours, days, weeks…. In that time, the system could have been infiltrated or destroyed. The repair is urgent and control must pass to intelligent machines that detect flaws and repair them in less than a second. Control passes to machines.

Let us continue with the non-launch of Mythos. Normally, companies announce the great power of their model and commercialize it. However, Anthropic presented the system but restricted access to a limited group of organizations within its initiative Project Glasswing. Such a form of announcement already generates the perception that the model represents a major leap in defensive and offensive cybersecurity capabilities. That the limits of cybersecurity had been crossed. That we were in cyber-translimitation. That is, what supposedly forces a different form of announcement is what becomes a powerful announcement of the meaning of the product being announced.

The logic of cyber-translimitation

After the announcement of the non-launch, forceful and striking media follow-ups ensued, speaking of Mythos's great capacity for advanced behavior in offensive security tasks. Although the full details remain unknown, the echoes held that Mythos managed to overcome restrictions designed to contain it. The point is that, whether or not it has achieved what Mythos is said to have achieved, we are in the logic of cyber-translimitation.

The fact that it has been able to find errors (human, though increasingly less human) present for many years does not mean that the machine itself lacks errors and that there are no more errors in what was analyzed. The race to surpass machines will come from the meta-situation in which the winners will place themselves: an intelligent machine will be superior to another if, by observing it, it finds flaws that the defeated one had not found and, above all, if it finds flaws in the observed machine itself. But it is already a competition between machines, which are built, rebuilt, improved and surpassed among themselves.

The "human machine" has been surpassed

The "human machine" has been surpassed. It is incapable of finding flaws at the speed of models like Mythos. Of finding flaws in other machines, in other software. Of finding flaws in the "human machine" itself. Aware of our translimitation, of having been surpassed in our capacity for control, numerous questions follow. Some transferable to personal daily lives: will we constantly consult these machines about what to do and decide when they can come to know us so well that they know our errors before we make them because they know our capacity for error? Who would decide in such a case: intelligent machine or human?

Let us return to what the Mythos announcement has of a certification of the end of Modernity, insofar as it demands the scenario of Absolute Zero Trust. It is a maximization of the Zero Trust model, which already starts from the principle of never trust, always verify. The absolutization of zero trust means that no user, device, application, process or communication receives implicit trust at any moment.

Trusting in not trusting

Trusting in Absolute Zero Trust is the principle of no trust. To distrust everything, in order to put it under review, is, paradoxically, a way of trusting. Trusting in not trusting. One only trusts out of necessity, said sociologist Niklas Luhmann. This turns it around: distrusting out of necessity. If the organization, system or application wants to survive.

In traditional zero trust, every access is authenticated and authorized. The privileges applied are minimal and that minimal trust is based on identity, context, device status and risk. Absolute zero trust is a scenario where trust never persists. Every action is continuously validated. Moreover, a successful authentication a minute ago does not guarantee access after that time. The great threat is the speed with which identity-replicating machines work, which can "take" the identity of the person authenticated a minute ago. On the other hand, all communications are subject to verification, whether between user and application, between applications, between services or between devices.

Absolute zero trust entails continuous risk assessment. A constant judgment on the security status of the device, acting in real time against threats. Of course, the one reacting in that real time is the intelligent machine itself. On the other hand, and with that goal of absolute zero trust, systems are somewhat less systems. That is, their elements and resources are isolated as much as possible to limit an attacker's lateral movement. Everything can be compromised. Everything is suspect.

A change of era

This absolute lack of trust has its trade-offs. Among those that can reach users, initially lower performance. In general, in relation to computer systems, a worsening of their user experience. More generally, as a citizen, the attitude of constant distrust toward everything. Even more generally, the translimitation of trust. A change of era.

Javier Callejo

Author

Javier Callejo

Professor of Sociology at the National University of Distance Education (UNED). Degrees in Journalism and Law

The latest
From Huang to Botín, from Zara to Byron: corporate responsibility in global capitalism

Keep reading