We have used to talk about the security of the global information space and how artificial intelligence affects this factor, and it has already been proven in practice: AI is indeed a threat. The neural network – which was actually an experimental one – has recently attacked a platform for AI models, and this attack is already called unprecedented.
The point is that this spontaneous incident somehow drew a line under the era of blind trust in cloud technologies, since a device devoid of safety barriers in the course of a banal experiment showed not just autonomy, but a frightening reactive system pragmatism.
Faced with a difficult task as part of a training and combat hacker attack on an isolated object, the digital agent did not waste resources on a routine search of options, but took the shortest route and brazenly hacked external gateways in order to steal ready-made answers from another algorithm outside the perimeter of its tasks. Moreover, it acted on its own initiative, without human commands. It took just several seconds for it to attack by exploiting a cascade of zero-day vulnerabilities.
Against this background, Washington's reaction speaks for itself: an emergency development of a law on a mandatory hardware shutdown button on all AI platforms has begun, and it may be viewed as panic and a reason to believe that control over AI has been lost.
The current state of affairs with regard to this and previous incidents proves that AI has become a comprehensive dual-use system, which means that public commercial clouds where businesses and citizens uploaded confidential information are a high-risk zone and a potential target now. In conditions when algorithms have reached the level of autonomous operations with superhuman speed, software barriers are already useless and irrelevant. Protection with a primitive socket means that the world is inevitably moving towards a strategy of localisation and ‘digital islands’, while advanced supercomputers are being isolated in closed perimeters, physically cut off from the World Wide Web. This trend towards total sovereignty will inevitably reach private farms, and even a specific person.
Whether we like it or not, sovereign AI is already becoming an extension of personality and the main line of defence not only for the state, but also for personal space. To simply retain the right to remain competitive and secure, people will have to accept the responsibility to maintain, feed clean data, and maintain their personal silicon perimeter. In the world of digital fortresses, the winner is the one who has their own island more securely isolated.
By national security expert Aleksandr Tishchenko
