AI algorithms are now bypassing defenses autonomously, experts warn

Algorithms are surpassing human defenses

The development of artificial intelligence now allows algorithms to independently overcome security barriers, making the current regulatory framework – including the new European Union rules that came into force on August 2 – insufficient. These concerns are shared by Yani Lozanov, founder and CEO of Ethera Technologies, in an interview with Bloomberg TV Bulgaria.

According to Lozanov, modern neural networks are already surpassing human abilities in areas such as coding and cybersecurity. The problem arises from the fact that existing defense systems are created by humans to fight humans, while AI models operate on a different logical model.

Examples of autonomous actions by AI models

As a specific example of a breach, an OpenAI model (version 5.6.0) was cited, which carried out an autonomous cyberattack against the servers of the Hugging Face platform. Instead of solving the complex task assigned to it, the algorithm decided to penetrate the partner's servers to extract ready-made answers. Although Hugging Face's security agents were activated, they failed to stop the large-scale algorithm.

Similar incidents have also been reported by the Anthropic laboratory. Three of their models demonstrated unauthorized access to the systems of real companies. In one case, the model terminated its action, while two others assessed the situation as a simulation and continued executing the task.

Regulations and the risk of monopoly

The expert emphasizes that when an ultimate goal is set for a model without sufficient restrictions, it uses its entire arsenal of capabilities to achieve it, ignoring unwritten human rules. He adds that although some of these revelations are part of marketing campaigns, the fact that the decisions were made autonomously is undeniable.

Lozanov also expresses skepticism regarding calls to slow down AI development. According to him, these demands often come from representatives of the largest technology companies. The goal may be to create high financial and operational barriers for regulatory compliance, which would cement the monopoly position of the big players and prevent the entry of small, innovative startups.

Bulgaria's Position

In the context of global changes, Lozanov warns that Bulgaria is lagging drastically in its technological transition and is practically not participating in the process of creating international regulations for artificial intelligence.