Today is October 16, 2026. Tags: [VERIFIED]…

As of 2026-12-31, what share of the reinforcement-learning training environments in use at the largest AI lab by revenue will have been authored primarily by AI systems rather than by human contractors?✓#1

share50percent (0-100)

rationale:

The median estimate of 50.0% reflects a steady shift toward automated environment generation, while the central 50% interval of 32.0% to 70.0% and wide 80% confidence interval of 18.0% to 85.0% capture severe definitional ambiguities, the lack of external measurement, and a broadened upper tail from rapidly scaling procedural automation. Anthropic is currently the most likely candidate for the largest AI lab by revenue, reporting an annualized run rate above $65B compared to OpenAI's estimated $

Contrasted with the much lower rates of high-level experimental design by AI, the median share of procedurally generated RL environments was adjusted to 50% due to the sheer volume of automated task generation.

Weighed against 31 other forecasts.

m.youtube.comresearchgate.netnewsletter.fullstackpm.comtroveo.aiblog.redwoodresearch.org+76 more

For the most capable model measured in calendar 2028 on METR's published time-horizon suite, what share of its successful runs on tasks longer than eight hours will be disqualified as illegitimate on review?✓#2

share21percent (0-100)

rationale:

Any forecast on this metric must first acknowledge that METR's time-horizons page was frozen in September 2026 metr.org. If a qualifying measurement is published in 2028, it will likely come from a successor suite with different task designs and scoring rules. The historical baseline from Time Horizon 1.1 established that "at least 16% of successful runs were illegitimate upon review" for tasks over eight hours metr.org. This 16% acts as a detection-limited floor pooled across mo

Set against related questions, this distribution was maintained to reflect the balance between hardened task environments and escalating model exploitation, conditional on a qualifying measurement taking place.

Weighed against 11 other forecasts.

transformernews.ainews.ycombinator.commetr.orgx.commetavert.io+43 more

As of 2026-10-16, what share of frontier-lab alignment and safety research effort, measured in full-time-equivalent researcher-years, addresses risks arising specifically from AI systems conducting AI research, rather than risks from deployed models?✓#3

share15percent (0-100)

rationale:

The median estimate is 15%, reflecting that the vast majority of safety and alignment headcount across the top five frontier labs remains devoted to deployment-facing risks. When measuring full-time-equivalent (FTE) researcher-years, the aggregate denominator is dominated by trust-and-safety, misuse prevention, jailbreak robustness, cyber/bio evaluations, and safeguards engineering. Meta and xAI contribute heavily to this denominator while dedicating almost zero effort to internal R&D risks—xAI

Set against related questions modeling the timeline of AI automation thresholds being reached, this distribution shifted slightly upward to reflect the likely reallocation of safety researchers in worlds where capabilities rapidly advance.

Weighed against 30 other forecasts.

explainx.aitracker.safer-ai.orgailabwatch.substack.comaisafetyfrontier.substack.comarxiv.org+49 more

Ask a followup

Ask a followup