METR proposes public assessments of frontier AI loss-of-control risk: three reviews on alignment, control and AI R&D acceleration

AI evaluation organization METR has proposed a public assessment program for frontier AI developers, aimed at giving the public real evidence about “loss-of-control risk” rather than verbal assurances. The program is built around three reviews. The first is alignment: reviewing unexpected incidents that have already occurred inside developers and checking whether the training process masked more serious misalignment. The second is control: assessing whether internal agents could cause serious harm without being detected, and whether existing monitoring and response are reliable. The third is AI R&D acceleration: quantifying how far AI automation is driving internal capability growth and algorithmic progress. METR writes a public report for each assessment, based on staff interviews and internal data. Developers may redact sensitive information and decide how open to be, but must disclose key limitations such as the scope of the assessment, what was redacted and when it is published. METR also stresses that the program only provides evidence: it is not a safety guarantee and does not amount to formal regulation, and it encourages developers to bring in more outside organizations and proactively disclose evidence. > Source: [read the original](https://metr.org/blog/2026-10-14-better-informing-the-public-about-loss-of-control-risks/). This text is a summary for reference only.