OpenAI Declares "Beginning of AGI Era" and Releases GPT-6

OpenAI has released a new model called "GPT-6 Astra." Greg Brockman, president of the company, has positioned this model's arrival as "the beginning of the AGI (Artificial General Intelligence) era," marking OpenAI's first official declaration of AGI's emergence.

AGI refers to a concept of artificial intelligence capable of performing a wide range of intellectual tasks like humans, not just specific tasks, and has long been considered a goal of AI research. OpenAI was established with the explicit aim of realizing AGI, and this statement demonstrates that internal assessment of that goal has undergone significant change. Nevertheless, what the phrase "AGI era" precisely means remains at a stage where external verification cannot yet be considered sufficient.

In terms of performance, according to OpenAI's announcement, GPT-6 Astra achieved state-of-the-art performance on benchmarks in mathematics, coding, and cybersecurity. Notably, in the cybersecurity domain, it reportedly independently discovered 2 zero-day vulnerabilities—previously unknown security flaws with potential for exploitation before disclosure—during testing that no one had discovered before. This is noteworthy as it demonstrates capability that could function in real-world security scenarios.

Furthermore, OpenAI has positioned GPT-6 Astra as its first model classified as "Critical" in its proprietary safety assessment framework. This means the model has been internally judged to carry a higher degree of risk than previous models. The specific criteria by which "Critical" was established as a safety assessment category and what countermeasures it entails remain unclear at this stage.

This development occurs within the context of intensifying competition across the AI industry as a whole. As multiple companies accelerate the development of high-performance models, OpenAI's definition of its own model using the term "AGI" can be understood as having not only a technological dimension but also an aspect of positional competition within the industry. Whether a model actually meets the criteria for AGI can only be discussed once accompanied by independent evaluation from researchers or third-party organizations.

The autonomous discovery of zero-day vulnerabilities represents a double-edged sword for the security field. If defensive actors leverage this capability, it could help with early detection of flaws that conventional methods would miss; conversely, if it reaches malicious actors, it carries the risk of triggering serious cyberattacks. The risk awareness underpinning this calculation likely informed OpenAI's decision to assign the "Critical" classification.

The release of GPT-6 Astra should be understood as a significant development signaling that artificial intelligence is transitioning from a research phase to one with tangible impact on real-world society. Going forward, how OpenAI's own safety assessment framework actually functions in practice, and how external researchers and regulatory authorities evaluate this model, will be points of significant interest.