By the end of 2027, AI systems will be able to complete tasks that take a skilled person weeks.
Reassessed
Consistency flag (unweighed consideration: METR baseline-time reliability, doubtful). Valid: added the baseline claim as a supports premise under the extrapolation argument, rewrote and re-evaluated the argument, and reassessed with explicit sensitivity arithmetic. Status supported unchanged; confidence 0.65 to 0.62, credence 0.75 to 0.70. Change is small; no dependents notified.
Reassessed: still Supported
verdict confidence 0.65 → 0.62 · credence 0.70
Structured and assessed
Reworded so the deadline is the end of 2027 and the hedge "could" becomes "will be able to". Added one named argument, extrapolating the time-horizon trend, linked to the existing post-2024 doubling-time quantity, and placed the Anthropic trend passage on it. Recorded MirrorCode (Epoch/METR) as an affirming instance. Assessed as supported, credence 0.75, importance 0.6, contestation 0.55.
Assessed Supported
verdict confidence 0.65 · credence 0.75
The forecast rests mainly on METR's measurements of AI "time horizons": the length of software tasks, measured by how long skilled humans take, that frontier agents complete with 50% success. In early 2026 the leading public models measured about 12 hours (Claude Opus 4.6, after a correction from 14.5), with stronger models put at 16 hours or more, and recent doubling times are estimated at roughly 3.5 to 4.5 months. If that growth continues even at the older seven-month pace, the 50% horizon passes a 40-hour working week during 2027, and at the recent pace it passes several weeks well before the end of the year. There is also direct evidence that the threshold is already crossed for some tasks. In April 2026, Epoch AI and METR's MirrorCode benchmark reported Claude Opus 4.6 fully reimplementing a 16,000-line program that engineers estimated would take a human 2 to 17 weeks. Those tasks are precisely specified and checkable, and they have no human baselines, so they show capability on a favourable kind of task, not on weeks-long work in general. The main uncertainties are about meaning rather than trend. The measured horizons are for software tasks at 50% reliability; 80% horizons are several times shorter, and messier, less well-specified work in other fields has not been shown to follow the same curve. METR's own suite was nearly saturated in 2026, so a weeks-long horizon may not be cleanly measurable by the end of 2027. Read as "frontier AI will be able to complete, with substantial reliability, some or typical software tasks that take a skilled person a week or more", the claim is well supported. Read as reliable performance across weeks-long tasks in general, it is much less certain. The most prominent statement of the forecast comes from Anthropic, whose own models are the subject of the evidence, though the underlying measurements are METR's and Epoch's.
Claim entered the graph