U.S. Agencies Face Aug. 1 Deadline to Set Classified Frontier AI Benchmarks

According to Axios, federal agencies have until Aug. 1 to establish a classified benchmarking process to assess the capabilities of frontier AI models, though the Financial Times reported the standards could arrive as soon as this week. Separately, when Fable 5 returned last week, Anthropic said it is building a standardized benchmark with Amazon, Google, Microsoft and other partners that focuses on the outcomes and impact of a jailbreak rather than whether one is possible. Axios also reported that testing lab Irregular released a new cyber benchmark in late June aimed at measuring whether AI models can execute offensive cyber tasks such as remote code execution, privilege escalation and reaching a restricted network, while other groups including Wiz and Vals AI are developing similar measures; Stanford’s 2026 AI Index warned that evaluations designed to remain difficult for years can be “saturated in months.”