In a step that represents a quantum leap in the world of artificial intelligence, OpenAI launched its new model, “GPT-5.4-Cyber”, which is a customized version of the pioneering “GPT-5.4” model, specifically designed to enhance cyber defense.
The launch comes at a time when cyber defense extends to protecting the physical world from cyber attacks targeting robots, smart machines, and the vital infrastructure of countries. However, amidst this technical momentum, a basic human angle emerges that addresses how this intelligent guard can gain the public’s trust instead of raising its fears.
The evolution of artificial intelligence towards cyber defense
Since the launch of the first GPT models, OpenAI has been focusing on general language and programming capabilities, but has moved towards developing specialized models in cyber defense in light of the increasing cyber threats that have become an existential threat to economies and governments.
GPT Cyber can predict attacks before they occur, and differs from previous models in that it is designed within a software environment that simulates complex hacking systems.
The new model is currently available through the TAC Trusted Access Program to a select and trusted group of experts, organizations and teams, reflecting a cautious approach that combines innovation and responsibility.
The release of GPT 5.4 Cyber comes about a week after Anthropic unveiled its pioneering model, Claude Mythos, which was able, in the pre-release testing phase, to uncover thousands of previously unknown security vulnerabilities in all major operating systems and major web browsers.
Cloud Mythos finds vulnerabilities that have survived decades of human security review and millions of automated tests, and is able to reproduce vulnerabilities and develop effective vulnerabilities on the first try in more than 83% of cases.
In one documented case, he exposed a 27-year-old vulnerability in OpenBSD, an operating system known for its security vulnerabilities.
Smarter guarding
The company designed the GPT 5.4 Cyber model to be less restrictive in responding to queries related to analyzing cyber attacks, allowing professionals to conduct security research that previous models rejected for safety reasons.
The model has the advantage of binary reverse engineering, which allows the analysis of compiled software – such as definition programs and commercial libraries – without owning the original source code, something that previously required human experts with very rare skills.
This technology is used to analyze malware, discover security vulnerabilities, compatibility between systems, and develop security skills.
The model analyzes massive code in real time, discovering zero-sum vulnerabilities that might be missed by the human eye or traditional tools, which is vital to combating complex supply chain attacks, such as the XZ Backdoor vulnerability that targeted Linux systems and shook the free software community.
According to OpenAI, the model – even in the initial testing stages – contributed to fixing many critical vulnerabilities, transforming it from an analytical tool into a security guard that works alongside human experts, suggests immediate solutions, and prevents malicious exploitation.
In test environments, the model helped security teams analyze compiled software that contained hidden vulnerabilities, similar to those used in previous attacks on industrial control systems.
Securing robotic systems
The current year is witnessing a historic transition from digital artificial intelligence to physical artificial intelligence, where cyber conflict is no longer just lines of software, but rather affects the heart of physical reality, such as factories, energy networks, and robotic systems.
This shift from digital AI to physical AI means that these systems are able to perceive and operate in complex physical environments to protect everyday life.
Industrial robots, self-driving cars, drones, and smart city management systems rely on hackable software, which makes hacking the artificial intelligence system that controls a surgical robotic arm or the high-speed train network a direct threat to lives, in addition to the loss of data.
Here, the role of GPT 5.4 Cyber emerges, which is capable of protecting cyber-physical systems that link computer components and physical processes and allow intelligent interaction between the digital and physical world.
The reverse engineering capabilities in GPT 5.4 Cyber allow for accelerating the detection and repair of such attacks on industrial robotic systems, as the new model analyzes vulnerabilities in these control systems and prevents attacks that turn them into an unintended weapon.
The new model can predict minute mechanical deviations resulting from a digital hack, and close the vulnerability within a short period of time before it turns into a tangible physical catastrophe.
Protecting national infrastructure
In the era of hybrid warfare, national infrastructures, such as power grids, transportation systems, water plants, and nuclear plants, have become a favorite target for cyberattacks.
For example, the Volt Typhoon group infiltrated US power grids for months, while hackers were able to tamper with the control systems of five US water and wastewater plants, leading to exceeding water levels in tanks and triggering false alarms.
In 2023, the hacking group Cyber Av3ngers succeeded in infiltrating programmable logic controllers in several American water and wastewater systems, manipulating the screens with the aim of causing disruption in operations.
In this field, GPT 5.4 Cyber provides preventive solutions by monitoring software used in industrial control systems, while turning threats into opportunities for early detection, especially with the expansion of the Cybersecurity Trusted Access Program to include the protection of critical infrastructure.
The model can detect malware hidden in industrial robotic systems before it leads to a disaster, as happened in previous attacks on power plants, as its capabilities extend beyond securing personal computers to securing critical infrastructure.
The company, which launched Codex Security as a means of discovering, verifying, and proposing solutions for security vulnerabilities, revealed that the application security agent powered by artificial intelligence has contributed to fixing more than 3,000 critical and high-risk security vulnerabilities in software databases since its launch.
But OpenAI is not alone in this field, as competition with Anthropic has created a new global market for sovereign security solutions.
The two companies seek to provide guarantees to governments that their infrastructure systems are immune from any external interference by relying on their artificial intelligence models, which turns this technology into a fundamental pillar of the concept of national security, especially in light of the current geopolitical tensions.
Gain the public’s trust
Regardless of the great technical capabilities, humans remain the most cautious link, especially with the emergence of the growing crisis of confidence, as people find it difficult to accept the idea of a machine making crucial decisions without human intervention.
GPT 5.4 Cyber’s ability to make immediate security decisions means excluding the human element from the decision-making process to ensure speed, which raises legal and ethical concerns about the transparency of these decisions and the ability to hold the algorithm accountable if it miscalculates the situation.
OpenAI recognizes this and its need to gain people’s trust, which has led it to restrict access to its Cybersecurity Trusted Access Program.
This approach is not just a security measure, but rather a message through which the company explains that it provides the capabilities of the smart guard to defenders only, not to attackers.
In designing GPT 5.4 Cyber, OpenAI relied on explainable artificial intelligence, meaning that the model does not make a defensive decision without providing an explanation in clear human language to security supervisors about why and how the action was taken.
This aims to cooperate with the human element as a strategic partner that gives ethical cover to automated decisions.
From a human perspective, the model highlights true stories about cybersecurity experts who were able to save vital systems thanks to its capabilities in initial tests, which restores confidence to the average citizen who fears superhuman artificial intelligence.
The company has also invested in grant programs and open source tools to enhance global capabilities, making technology a tool for the greater good, not a corporate monopoly.
However, the picture is not without its shadows, as it is human trust built on transparency and accountability that determines the true success of this model, because the capabilities of the model may be exploited if they fall into untrusted hands, and over-reliance on artificial intelligence may create new vulnerabilities if the model itself is attacked.
Issues of privacy and ethics also raise questions about who determines the legitimate defender in a world full of conflicts, especially with the increase in attacks powered by artificial intelligence, such as deepfakes.
In conclusion, OpenAI has succeeded in transforming defense with artificial intelligence from a mere marketing slogan into a real software tool capable of writing defense instructions and correcting vulnerabilities immediately, marking the beginning of a new era in which artificial intelligence merges with physical reality to protect humanity.
