A METR report with deep access at Anthropic indicates the company has crossed its Responsible Scaling Policy threshold for "dramatically accelerated progress due to AI," defined as roughly doubling the rate of capabilities progress compared to the fastest sustained pre-AI pace, with AI automation as the primary driver. Between October 2025 and October 2026, progress measured by the internal Anthropic Effective Compute Index (AECI) ran about twice as fast as the prior year; the report authors attribute roughly 80% of the current speedup to AI, a contribution that has grown over time. Because capabilities equal compute multipliers times training compute, doubling overall progress requires compute multipliers to grow ~2.5–3x faster. Modeling researcher labor and experiment compute as equally important Cobb-Douglas inputs implies a ~4–9x labor uplift to achieve that multiplier acceleration; accounting for diminishing returns and the need to sustain the new pace suggests a current uplift of ~10x, aligning with Anthropic staff reports of ~12x. This uplift functions as a serial time-warp—researchers effectively think and act ten times faster—rather than parallel scaling. The report found no clear bottlenecks or steep diminishing returns to prevent compounding acceleration; if smooth, doubling times could shrink from six months to three months to six weeks, naively reaching extreme capability within a year. The previous RSP threshold (AI R&D-5) required ASL-4 security (protection against state-level model-weight theft), but both OpenAI and Anthropic have suffered breaches by amateur groups. The author argues the industry lacks the security, monitoring, alignment verification, and governance infrastructure to handle an intelligence explosion responsibly, and criticizes the redaction of most supporting evidence, which forces reliance on trust rather than independent verification.