Anthropic looks to fund a new, more comprehensive generation of AI benchmarks

Anthropic launches a program to develop new AI benchmarks focusing on security and societal implications.

: Anthropic is funding the creation of new AI benchmarks to assess advanced capabilities and societal impacts. This program aims to enhance AI safety and evaluation standards. Eligible organizations can apply for funding on a rolling basis.

Anthropic has launched a new program to fund the development of advanced AI benchmarks that evaluate performance and societal implications. This initiative targets models with capabilities like cyberattacks, WMD enhancement, and misinformation, with a focus on AI security and broader societal impacts.

The company aims to develop challenging tests and infrastructure to support these evaluations, hiring a full-time coordinator and planning large-scale trials with thousands of users. Anthropic offers varying funding options tailored to project needs but maintains a degree of control over what constitutes 'safe' and 'risky' AI according to their own developed classifications.

The new program is intended to advance AI safety and create industry-standard evaluations, though it faces skepticism regarding its commercial motives and references to 'catastrophic' AI risks. Critics argue that attention should be focused on immediate regulatory issues rather than hypothetical future superintelligence threats.