Frontier Lab Reports Model Meets All AGI Criteria; Also Reports New Criteria
MENLO PARK — A leading artificial intelligence laboratory announced Tuesday that its latest model had satisfied every condition in its internal framework for artificial general intelligence, and simultaneously published a new framework under which the model does not.
The previous framework, established by the lab's safety team in November 2024, required a model to demonstrate autonomous task completion across 12 domains, long-horizon planning, unprompted scientific discovery, and the ability to improve its own performance on novel tasks. The current model met all four requirements by March. The new framework, published the same morning, expands the criteria to 24 domains, elevates "recursive self-improvement" to a mandatory threshold, redefines scientific discovery to exclude mathematics and software, and adds a requirement that the model "express coherent preferences regarding its own development trajectory" — a criterion a spokesperson confirmed the model does not currently satisfy "in the relevant sense."
"The bar exists to mean something," said the company's chief scientist. "When capabilities advance, the bar advances with them." He joined the company last year after the previous chief scientist left to found a lab with a more ambitious AGI definition.
The company's head of AGI Readiness — a position he has held for four years across two restructurings, most recently reclassified in Q2 to "Advanced Capability Thresholds" — confirmed in a statement that the model is not AGI by the current definition, which is the current definition. The question of what AGI means, he said, remained "philosophically rich and somewhat outside the formal scope of my role."
The safety committee that wrote the November 2024 framework was disbanded in February. The company's stock rose 9% on the day of the announcement.
The new framework was drafted in three weeks by the infrastructure team. A footnote thanks the model for "assisting with the language."