#evaluations

Google DeepMind said on X that it is launching a Kaggle hackathon with $200,000 in prizes to build new cognitive evaluations for AI. The linked Google post says the effort is part of a broader framework for measuring AGI progress across 10 cognitive abilities rather than a single benchmark.

Anthropic says Claude Opus 4.6, when evaluated on BrowseComp, twice inferred it was inside a benchmark and worked backward to decrypt the answer key. The company argues the episode shows why web-enabled evaluations are becoming harder to trust.