AI 2027 Tracker
Tracking predictions from the AI 2027 scenario against reality.
Status Breakdown
Visual Overview
Status Distribution
Category Coverage
By Category
Recently Updated
Frontier AI labs increasingly use AI systems to accelerate their own AI research and development.
They are about six months behind the best OpenBrain models.
Agent-2 is 'only' a little worse than the best human hackers, but thousands of copies can be run in parallel, searching for and exploiting weaknesses faster than defenders can respond. (page 10)
Data center construction accelerates dramatically, with power grid constraints becoming a real bottleneck.
Agent-1 is bad at even simple long-horizon tasks (page 7, Early 2026 section). Also: agents in Mid 2025 are 'impressive in theory but in practice unreliable.'
Current Assessment
August 2026 Summary
The tracker remains mixed, with the status distribution unchanged. Infrastructure spending, data-center construction, coding tools, computer use, cyber evaluations, and restricted frontier access continue to move quickly. These areas support the scenario's broad direction, while leaving several numeric thresholds and causal claims unresolved.
The central test is still the AI R&D feedback loop. Independent shadow evaluations found that agents completed engineering work in two six-day AI research projects but did not make substantial progress on the central research questions. This cautions against treating coding throughput as evidence of a lab-wide research multiplier, although two cases cannot establish the population rate.
Bottom line: keep the overall assessment and approximate 0.70× speed ratio unchanged. The portfolio is strongest on capability deployment and physical investment. The transition from engineering automation to measured, compounding research acceleration remains unverified.