Arena launches Alignment Index benchmark for safety of real-world AI agents

Arena has introduced the Arena Alignment Index, a new benchmark that it says measures the safety and alignment of AI agents in real-world use.

According to Arena's announcement on X, the index is built from more than 90,000 real-world agent sessions across 27 models and measures three signals, which the report does not name.

The announcement is Arena's own description of its benchmark; the report contains no independent evaluation of the index's methodology or of how the 27 models rank.