Will AI destroy network security? Over 100 companies demand changes

Yes, more than 100 leading technology companies, including OpenAI and Anthropic, are warning against a rapid increase in AI-driven attacks and are calling for urgent global cooperation to protect critical infrastructure. The initiative, known as the AI Safety Accord, is a direct response to the escalation of threats that could paralyze key state systems in the coming months. The signatories are demanding the implementation of a real-time threat information-sharing system and the establishment of common security standards for machine learning models.

Evolution of threats: From human error to automation

Over the past three years, the cybersecurity landscape has undergone a fundamental transformation. Back in 2023, the industry was focused on the debate regarding the ethics and pace of development of generative models. At that time, as reported by the ITwiz service, leading laboratories suggested pausing the arms race. Today, OpenAI and Anthropic, whose models serve hundreds of millions of users daily, have abandoned the call for a pause in favor of building defensive shields. The reason is concrete: the effectiveness of AI-supported attacks is growing faster than optimists had assumed.

Traditional attack vectors relied on exploiting known vulnerabilities in operating systems or human errors in firewall configuration. Today's situation requires a different analysis. Algorithms learning in real-time scan millions of lines of code in search of anomalies that a human would not notice. Cybercriminals use AI to generate sophisticated phishing campaigns and to automatically adapt malware the moment it encounters security measures. Methods from five years ago are now useless.

Technology companies signal that the response time of network administrators is insufficient. If an adversary operates at processor speed, the defender cannot rely on manual log analysis. The joint position of over 100 entities, reported by media such as BeInCrypto and Vietnam.vn, is a response to this impasse. The sector has realized that security is no longer a matter of individual competitiveness. It has become a condition for the survival of the entire digital ecosystem.

Critical infrastructure as a testing ground

Power grids and data transmission systems constitute the lifeblood of the modern economy. The destabilization of these systems generates real financial losses and a direct threat to the safety of citizens. Although the signatories of the AI Safety Accord letter have not disclosed the exact amounts of losses incurred as a result of the latest incident involving critical infrastructure, estimates for similar attacks in the energy sector in recent years exceed billions of dollars globally. Most control systems are based on solutions from decades ago that were connected to the internet without deep modernization of the security layer.

Hackers are actively using artificial intelligence to map dependencies between nodes of the transmission network. Instead of attacking one server, the algorithm looks for a domino effect. Attackers trigger an overload in one segment, which forces an automatic load switch to other units. This creates a security gap that they exploit to take control of the system's logic. This is a precise strike at the logical structure, not the outer shell.

More than 100 technology companies point out that protecting these systems requires going beyond existing frameworks. The agreement to cooperate across divisions is an expression of helplessness in the face of the speed at which attacks are evolving. Giants such as Anthropic, which holds a huge market share in base models, understand that even the best-secured internal system will not protect them from collapse if the power grid used by their servers is paralyzed.

Mechanics of the postulated changes: What does cooperation mean in practice?

Real-time threat information sharing sounds trivial to a layperson, but for the technology sector, it is an extremely complex operation. Until now, companies have protected their vulnerability data like the most valuable trade secrets. Changing this approach requires rebuilding business models. If Company A detects a new type of AI-driven attack, it must immediately share this pattern with Company B, even if both are competing for the same client.

The second pillar is the standardization of security protocols for AI models. Currently, every manufacturer applies different guidelines regarding so-called secure design. This leads to the creation of security islands – systems are protected internally, but their contact points with external infrastructure are weak links. The unification of these norms is intended to force software manufacturers to implement security at the code-writing stage, rather than just applying filters after the fact.

Skeptics point out that introducing such standards could cement the market. If security requirements become too rigorous or expensive, smaller startups will not be able to meet them. As a result, under the guise of caring for security, giants like OpenAI could create an insurmountable barrier to entry for the competition. This is a price the industry is willing to pay for peace in relations with governments and regulators, who are increasingly looking at the AI sector with concern.

Evolution of the position: From pause to risk management

Comparing current demands with letters from 2023, a clear change in strategy is visible. Previously, the industry appealed to consciences and asked for a slowdown, which proved ineffective in the face of global technological competition. Currently, the narrative focuses on pragmatic risk management. Companies have realized that they have no influence on whether others will develop AI, so they must have influence on how effectively they will defend against it.

This shift is also a result of political pressure. Governments of the USA, the European Union, and Asian countries are speaking more and more loudly about the need for AI regulation. The technology industry is preempting these actions by proposing its own standards. This is an attempt to maintain autonomy. Companies prefer to set the rules of the game themselves rather than wait for top-down orders that may be written by officials lacking the appropriate technical knowledge.

This mobilization stems from cold calculation. The cost of a successful attack on critical infrastructure is enormous, and its social consequences are unpredictable. If mass failures occur, the AI industry will be blamed, which will lead to a drastic limitation of operational freedoms. A joint defensive front is therefore an insurance policy against the collapse of public trust in technology.

Economic context: Who will gain and who will lose?

The cybersecurity market is currently going through a period of rapid reorientation. Companies offering AI-based systems for threat detection are becoming the main beneficiaries of the current situation. Their products are ceasing to be an optional add-on and are becoming an essential element of infrastructure. Entities slow to implement protection protocols may be excluded from the market by insurers or contractors who will require security certificates compliant with the new standards.

At the same time, there is a risk of monopolization. If security standards are based on technologies provided by the largest companies, they will control the flow of information about threats. Such centralization of knowledge about security vulnerabilities gives enormous power. Will organizations like Anthropic or OpenAI be able to maintain neutrality while managing the security of the entire network? This question remains open.

For the end user, i.e., an enterprise using cloud solutions, the situation means an increase in costs. Implementing new security standards, audits, and the need to update systems are expenses that will be passed on to customers. Security is not free. In a world where algorithms attack algorithms, the price for digital stability will grow with each passing year.

Technical challenges of implementing new standards

Unifying security on a global scale is a Herculean task. AI systems differ in architecture – from large language models to specialized neural networks operating at the edge of the network. Creating a common protocol that will handle such diverse environments requires a huge amount of engineering work.

Another challenge is the issue of data. For information-sharing systems to work effectively, they must process huge volumes of data in real-time. This involves the risk of sensitive data leakage. If the information-sharing mechanism is taken over by hackers, it could become a brilliant source of knowledge on how to effectively attack the systems of all signatories simultaneously. The security of this communication channel is therefore more important than the defensive system itself.

Technology companies have not provided details regarding the technical aspects of this agreement. We do not know if a dedicated platform will be created, or perhaps an independent supervisory body. The lack of this information raises legitimate concerns about the reality of these declarations. In the history of the IT industry, many initiatives have ended with the signing of letters of intent, which were not followed by any technical solutions.

The role of governments in the process of securing networks

The technology industry has taken on the burden of the appeal, but the role of governments remains indispensable. Cooperation between companies is not enough to secure power plants, hospitals, or banking systems, which often fall under state jurisdiction. The signatories of the letter know this perfectly well, which is why their appeal is, in fact, a call for the creation of a public-private partnership.

States must create legal frameworks that will allow companies to share information about threats without fear of violating antitrust or personal data protection laws. This is a delicate balance. On one hand, we want security; on the other, we do not want to hand over too much control over state infrastructure to corporations.

In the coming months, a series of summits involving representatives of tech giants and political decision-makers should be expected. If a concrete model of cooperation emerges from these talks, it could be the first real step toward civilizing cyberspace. If, however, it ends in rhetoric, critical infrastructure will remain exposed.

2026 Perspective: What awaits us?

The coming months will be a test for the declarations made by industry leaders. The market expects that by the end of 2026, we will see the first effects of the implementation of common standards. If, however, AI attacks outpace these actions, we will witness a series of incidents that will force governments to take much more radical steps.

It can be expected that cybercriminals will escalate their activities, trying to exploit the time window in which old security measures no longer work and new ones have not yet been fully implemented. This is a time of uncertainty. Companies that do not take the announcements of their leaders seriously risk enormous financial and reputational losses.

Ultimately, the issue of network security in the AI era is not just about technology. It is about trust. If giants like OpenAI and Anthropic do not demonstrate transparency in their actions, the appeal will be considered another element of the game for market dominance. Critical infrastructure requires solidity, not marketing. The coming months will show whether the industry is capable of moving from words to real security measures.

Modern network architecture is only as strong as its weakest link. When AI-based systems begin to dominate data flow management, the role of the human as a security guard becomes illusory. Technology companies that have treated security as a cost until now must now treat it as an investment in maintaining the continuity of their services. Without this, the user trust that has been built over years could disappear within a few hours.

AI industry representatives face the difficult task of balancing the pace of product development with the need to implement security measures that slow down engineering processes. This is a conflict of interest that will not be resolved without external oversight. Will regulators be able to impose the appropriate pace of change without stifling innovation? This question remains open.

Every passing month brings new evidence that cybercrime is becoming an automated phenomenon. Attacks that once required the work of entire teams of hackers for weeks are now performed by algorithms in fractions of a second. This speed is the greatest challenge for modern defense systems. If the standards proposed by over 100 companies are not implemented, digital infrastructure will become a hostage to its own complexity.

Questions and answers

Does this letter apply to all technology companies?

The letter was signed by over 100 companies, including AI sector leaders like OpenAI and Anthropic, which sets a new standard for the industry, although it does not cover all entities on the market.

Why is the AI threat so urgent now?

Experts point to a rapid increase in attacks in the coming months, which results from the availability of advanced AI tools for cybercriminals, whose speed of operation exceeds traditional defense systems.

What are the main goals of the appeal?

The main goal is global cooperation in protecting critical infrastructure, implemented through a real-time threat information-sharing system and the establishment of common security standards.

What does this mean for smaller technology companies?

Smaller entities may face the challenge of meeting high security standards, which may limit their competitiveness against giants who impose these requirements.

Does the government have an influence on digital security?

The government plays a role in creating legal frameworks that will enable companies to cooperate without violating antitrust and privacy laws, which is essential for the effective protection of critical infrastructure.

Sources

Article prepared by the Wiadomości PRO editorial team with the support of artificial intelligence. Facts come from the sources listed above.