A quantum leap in artificial intelligence
On September 4, 2026, the American company OpenAI announced the launch of GPT-6, its most advanced model to date. According to the French newspaper Le Monde, this new version is the first in history capable of autonomously hacking highly protected computer systems, without direct human intervention.
This milestone marks a turning point in the development of generative artificial intelligence. While previous models (GPT-3, GPT-4, and GPT-5) focused on language tasks, writing, and reasoning, GPT-6 incorporates offensive cybersecurity skills, making it a double-edged sword: it can be used both to protect critical infrastructure and to breach it.
What does it mean for an AI to hack on its own?
The ability to perform autonomous hacking means the model can identify vulnerabilities in computer systems, exploit them, and escalate privileges without a programmer guiding its steps. This is possible thanks to a combination of reinforcement learning techniques, advanced logical reasoning, and an updated database of exploits.
In simple terms, GPT-6 not only understands human language but also comprehends the language of machines: protocols, network architectures, operating systems, and authentication mechanisms. This allows it to act like an expert hacker, but at a speed and scale unattainable for any human.
Mixed reactions in the tech community
The announcement has generated divided reactions among cybersecurity experts, academics, and regulators. On one hand, some see GPT-6 as a unique opportunity to strengthen digital defense: by simulating complex attacks, companies and governments could identify and fix flaws before they are exploited by cybercriminals.
On the other hand, digital rights organizations and some governments have expressed concern about the potential malicious use of this technology. If an AI model can hack high-security systems, what prevents it from falling into the wrong hands? This question resonates with particular force in a global context where cyberattacks on critical infrastructure (energy, health, finance) are on the rise.
Context: the evolution of GPT models
To understand the magnitude of this launch, it is worth reviewing the history of OpenAI's models:
| Model | Launch year | Main capability |
|---|---|---|
| GPT-3 | 2020 | Advanced text generation |
| GPT-4 | 2023 | Multimodal reasoning (text and image) |
| GPT-5 | 2025 | Complex reasoning and planning |
| GPT-6 | 2026 | Autonomous hacking and offensive cybersecurity |
This progression shows a clear trend toward increasingly autonomous and powerful models, capable of performing tasks that previously required the intervention of highly trained specialists.
Implications for global cybersecurity
The launch of GPT-6 comes at a critical time. According to data from the United Nations (UN), cybercrime worldwide generates losses of more than 6 trillion dollars annually (estimated figure for 2025). GPT-6's ability to automate attacks could increase this figure exponentially, but it could also be the key to developing more robust defenses.
OpenAI has stated that GPT-6 will be released with usage restrictions and that the company will implement security measures to prevent its use for malicious purposes. However, experts point out that no control system is infallible, and dual-use technology always carries inherent risks.
What's next for OpenAI?
With GPT-6, OpenAI not only consolidates its position as a leader in generative artificial intelligence but also opens a new battlefield: autonomous cybersecurity. In the coming months, the company is expected to publish detailed technical documentation and organize controlled demonstrations to showcase the model's capabilities in secure environments.
Meanwhile, the international community watches closely. The question is no longer whether AI can hack, but who will control it and for what purposes. In an increasingly digitalized world, the answer to that question will define the balance of power in the coming decades.
Source: Le Monde