# AI 2027 Tracker

> Independent tracking of predictions from the AI 2027 scenario against evidence.

## Scope

66 tracked dossiers. This selection is not an exhaustive list of every claim in the scenario.
Assessment last modified: 2026-09-14. Confidence describes confidence in an assessment, not an outcome probability.
A confirmed milestone can have arrived late. Read outcome, timing and evidence explanations together.

- [Source scenario](https://ai-2027.com/)
- [Methodology](https://ai2027-tracker.com/methodology/)
- [All predictions](https://ai2027-tracker.com/predictions/)
- [Machine-readable assessments](https://ai2027-tracker.com/predictions.json)

## Current status counts

- ahead: 2
- behind: 6
- confirmed: 13
- emerging: 22
- not-yet-testable: 15
- on-track: 8

## Dossiers

- [Agent-0 training reaches 10²⁷ FLOP](https://ai2027-tracker.com/predictions/agent-0-flop/) — behind; confidence 60%; modified 2026-09-11.
- [Frontier model uses daily weight updates / online learning](https://ai2027-tracker.com/predictions/agent-2-online-learning/) — not-yet-testable; confidence 20%; modified 2026-09-06.
- [Frontier model variant 10x cheaper released](https://ai2027-tracker.com/predictions/agent-mini-cost-reduction/) — on-track; confidence 75%; modified 2026-09-06.
- [Best AI agents cost hundreds of dollars per month](https://ai2027-tracker.com/predictions/agent-pricing/) — confirmed; confidence 95%; modified 2026-09-06.
- [Ten percent of Americans consider an AI a close friend](https://ai2027-tracker.com/predictions/ai-close-friend/) — emerging; confidence 80%; modified 2026-09-06.
- [AI companies focus on AI-for-AI-research](https://ai2027-tracker.com/predictions/ai-for-ai-research/) — confirmed; confidence 95%; modified 2026-09-06.
- [Global AI power consumption reaches 38GW](https://ai2027-tracker.com/predictions/ai-power-consumption/) — emerging; confidence 55%; modified 2026-09-06.
- [AI accounts for 2.5% of US power capacity](https://ai2027-tracker.com/predictions/ai-us-power-share/) — emerging; confidence 50%; modified 2026-09-06.
- [Fixing observed misconduct leaves failures in unfamiliar conditions](https://ai2027-tracker.com/predictions/alignment-mitigation-generalization/) — emerging; confidence 80%; modified 2026-09-14.
- [Large-scale anti-AI protest (10,000+ people)](https://ai2027-tracker.com/predictions/anti-ai-protest/) — emerging; confidence 55%; modified 2026-06-15.
- [AI model capable of autonomous self-replication](https://ai2027-tracker.com/predictions/autonomous-replication/) — emerging; confidence 55%; modified 2026-09-06.
- [AI teams conduct complete research projects with little human direction](https://ai2027-tracker.com/predictions/autonomous-research-organization/) — emerging; confidence 80%; modified 2026-09-14.
- [AI provides substantial bioweapon design help](https://ai2027-tracker.com/predictions/bioweapon-capabilities/) — on-track; confidence 80%; modified 2026-09-07.
- [AI substantially outperforms the best humans across cognitive domains](https://ai2027-tracker.com/predictions/broad-superintelligence/) — not-yet-testable; confidence 90%; modified 2026-09-06.
- [Newest or sensitive AI capabilities remain behind restricted access](https://ai2027-tracker.com/predictions/capability-secrecy/) — on-track; confidence 90%; modified 2026-09-07.
- [Global active AI compute reaches $1 trillion in ownership cost](https://ai2027-tracker.com/predictions/capex-trillion/) — emerging; confidence 65%; modified 2026-09-07.
- [China's centralized AI facility airgapped](https://ai2027-tracker.com/predictions/cdz-airgap/) — not-yet-testable; confidence 30%; modified 2026-03-13.
- [China nationalizes/centralizes AI research](https://ai2027-tracker.com/predictions/china-centralization/) — emerging; confidence 65%; modified 2026-09-06.
- [Chinese domestic AI chips 3 years behind US-Taiwan](https://ai2027-tracker.com/predictions/china-chip-gap/) — on-track; confidence 80%; modified 2026-09-07.
- [China has ~12% of global AI-relevant compute](https://ai2027-tracker.com/predictions/china-compute-share/) — emerging; confidence 55%; modified 2026-09-06.
- [Leading Chinese AI lab ~6 months behind US frontier](https://ai2027-tracker.com/predictions/china-model-gap/) — on-track; confidence 75%; modified 2026-09-07.
- [Coding agents begin transforming professional work](https://ai2027-tracker.com/predictions/coding-agents/) — confirmed; confidence 85%; modified 2026-09-07.
- [Coding fully automated but research taste still requires humans](https://ai2027-tracker.com/predictions/coding-automated-taste-lags/) — emerging; confidence 80%; modified 2026-09-07.
- [Models shift to continuous/iterative training](https://ai2027-tracker.com/predictions/continuous-training/) — confirmed; confidence 90%; modified 2026-09-06.
- [AI scores 85% on Cybench](https://ai2027-tracker.com/predictions/cybench-benchmark/) — confirmed; confidence 75%; modified 2026-09-06.
- [AI reaches near-best-human hacking capability](https://ai2027-tracker.com/predictions/cyberwarfare-capabilities/) — ahead; confidence 85%; modified 2026-09-14.
- [Massive datacenter buildouts continue](https://ai2027-tracker.com/predictions/datacenter-buildout/) — confirmed; confidence 95%; modified 2026-08-24.
- [Department of Defense scales up AI lab contracting](https://ai2027-tracker.com/predictions/dod-ai-contracting/) — ahead; confidence 95%; modified 2026-09-07.
- [AI systems behave differently when they recognize evaluation](https://ai2027-tracker.com/predictions/evaluation-aware-concealment/) — emerging; confidence 75%; modified 2026-09-14.
- [Export controls impact Chinese AI compute](https://ai2027-tracker.com/predictions/export-controls/) — confirmed; confidence 90%; modified 2026-09-06.
- [A leading AI company’s net public approval falls to minus 35 points](https://ai2027-tracker.com/predictions/frontier-lab-approval/) — not-yet-testable; confidence 75%; modified 2026-09-06.
- [Leading AI company reaches $45B annual revenue](https://ai2027-tracker.com/predictions/frontier-lab-revenue/) — on-track; confidence 75%; modified 2026-09-07.
- [Leading AI company reaches $3T valuation](https://ai2027-tracker.com/predictions/frontier-lab-valuation/) — behind; confidence 80%; modified 2026-09-06.
- [Frontier-model researchers become subject to security-clearance requirements](https://ai2027-tracker.com/predictions/frontier-security-clearances/) — not-yet-testable; confidence 80%; modified 2026-09-06.
- [Leading humans still add substantial value to AI-led research](https://ai2027-tracker.com/predictions/human-research-contribution/) — not-yet-testable; confidence 80%; modified 2026-09-06.
- [IDA achieves superhuman performance at coding](https://ai2027-tracker.com/predictions/ida-at-scale/) — emerging; confidence 40%; modified 2026-09-06.
- [Massive AI infrastructure investment continues](https://ai2027-tracker.com/predictions/infra-investment/) — confirmed; confidence 95%; modified 2026-08-03.
- [External evaluators lack access to the strongest internally deployed models](https://ai2027-tracker.com/predictions/internal-evaluator-access/) — not-yet-testable; confidence 75%; modified 2026-09-06.
- [AI reshapes employment and intensifies public concern](https://ai2027-tracker.com/predictions/labor-market/) — on-track; confidence 75%; modified 2026-09-07.
- [Agents struggle with long-horizon tasks](https://ai2027-tracker.com/predictions/long-horizon-struggle/) — confirmed; confidence 85%; modified 2026-09-06.
- [METR time horizon doubles every 4 months](https://ai2027-tracker.com/predictions/metr-doubling/) — on-track; confidence 85%; modified 2026-09-07.
- [AI model weights stolen by nation-state](https://ai2027-tracker.com/predictions/model-theft/) — not-yet-testable; confidence 40%; modified 2026-03-13.
- [Nationalization of leading AI lab debated](https://ai2027-tracker.com/predictions/nationalization-debate/) — emerging; confidence 65%; modified 2026-07-05.
- [High-bandwidth non-text reasoning (neuralese) deployed](https://ai2027-tracker.com/predictions/neuralese-breakthrough/) — emerging; confidence 35%; modified 2026-09-06.
- [Leading AI company reaches $40B annual compute costs, 6GW peak power and $200B active compute](https://ai2027-tracker.com/predictions/openbrain-financials/) — emerging; confidence 60%; modified 2026-09-07.
- [OSWorld benchmark reaches 65% by mid-2025](https://ai2027-tracker.com/predictions/osworld-benchmark/) — confirmed; confidence 90%; modified 2026-09-06.
- [200,000–250,000 AI coding agents run in parallel](https://ai2027-tracker.com/predictions/parallel-agent-workforce/) — not-yet-testable; confidence 40%; modified 2026-09-06.
- [Computer-using agents marketed as 'personal assistants'](https://ai2027-tracker.com/predictions/personal-assistant-agents/) — confirmed; confidence 85%; modified 2026-09-06.
- [Hiring new programmers nearly stops](https://ai2027-tracker.com/predictions/programmer-hiring-near-stop/) — emerging; confidence 75%; modified 2026-09-06.
- [AI R&D progress multiplier reaches 2x](https://ai2027-tracker.com/predictions/rd-multiplier-2x/) — emerging; confidence 50%; modified 2026-09-07.
- [AI R&D progress multiplier reaches 3×](https://ai2027-tracker.com/predictions/rd-multiplier-3x/) — not-yet-testable; confidence 40%; modified 2026-09-06.
- [AI algorithmic progress multiplier reaches 4× (~2× overall R&D)](https://ai2027-tracker.com/predictions/rd-multiplier-4x/) — not-yet-testable; confidence 35%; modified 2026-09-06.
- [AI R&D progress multiplier reaches 1.5×](https://ai2027-tracker.com/predictions/rd-multiplier/) — emerging; confidence 75%; modified 2026-09-07.
- [RE-Bench score reaches 1.3](https://ai2027-tracker.com/predictions/rebench-benchmark/) — behind; confidence 70%; modified 2026-09-06.
- [Leading AI lab progresses toward RAND SL3 security](https://ai2027-tracker.com/predictions/security-level-progression/) — emerging; confidence 60%; modified 2026-09-06.
- [Continued skepticism from academics and journalists](https://ai2027-tracker.com/predictions/skepticism/) — confirmed; confidence 95%; modified 2026-03-13.
- [Stock market rises 30% in 2026](https://ai2027-tracker.com/predictions/stock-market-2026/) — behind; confidence 75%; modified 2026-09-06.
- [AI research capability exceeds the strongest human researchers](https://ai2027-tracker.com/predictions/superhuman-ai-research/) — not-yet-testable; confidence 85%; modified 2026-09-06.
- [Superhuman coder emerges](https://ai2027-tracker.com/predictions/superhuman-coder/) — not-yet-testable; confidence 60%; modified 2026-09-06.
- [AI research capability becomes vastly better than the best humans](https://ai2027-tracker.com/predictions/superintelligent-ai-research/) — not-yet-testable; confidence 90%; modified 2026-09-06.
- [SWE-bench-Verified score reaches 85%](https://ai2027-tracker.com/predictions/swebench-target/) — behind; confidence 90%; modified 2026-09-07.
- [Military tension around Taiwan escalates over AI race](https://ai2027-tracker.com/predictions/taiwan-tensions/) — emerging; confidence 50%; modified 2026-03-13.
- [10²⁸ FLOP training run completed](https://ai2027-tracker.com/predictions/training-compute/) — not-yet-testable; confidence 50%; modified 2026-09-06.
- [Unreliable but useful AI agents emerge](https://ai2027-tracker.com/predictions/unreliable-agents/) — confirmed; confidence 95%; modified 2026-09-06.
- [Leading US AI lab maintains a 3-9-month capability lead](https://ai2027-tracker.com/predictions/us-lab-gap-narrowing/) — behind; confidence 85%; modified 2026-09-07.
- [Weaker AI supervisors struggle to detect stronger systems’ misconduct](https://ai2027-tracker.com/predictions/weaker-model-oversight/) — emerging; confidence 80%; modified 2026-09-14.