What Is AGI?
The two defining characteristics usually attached to AGI are the ability to self-teach and human-like cognitive flexibility across a wide range of domains. There's no agreed-upon benchmark for when a system would qualify, and real disagreement among AI labs and researchers about how close current frontier models actually are to it.
Today's most capable models can perform impressively across many tasks without meeting the bar of self-directed, general-purpose learning that AGI implies. As a benchmark, it matters less for near-term security planning than the concrete capabilities, like autonomy and tool use, that current models already have.
Why It Matters
- The term gets used loosely in industry conversation, worth distinguishing "AGI" the aspiration from "increasingly capable current models" the reality when evaluating actual risk.
- Security planning is better anchored to what a system can actually do today (call tools, act autonomously, access data) than to a still-undefined future capability.
