AGI
Will AI Models Become Too Powerful?
Don't just read what happened. See what could happen next.
Prediction
Capability keeps rising into economically transformative territory while governance remains fragmented — “too powerful” becomes a political judgment more than a single technical threshold.
- Confidence
- 52%
- Horizon
- 24–60 months
- Impact
- High
- Direction
- Increasing
Prediction changed +7 points in 21 days
Short answer
Models already exceed humans on many tests and still fail oddly elsewhere. The past year’s debate is less “can they?” and more “who controls deployment, and for what?” Power concentrates in a few labs with opaque training stacks.
Why this question matters
Perceptions of excessive power drive licensing, export rules, nationalization pressure, and public backlash risk.
What's happening now?
Jagged superhuman performance
Gold-medal level math alongside basic perceptual failures fuels unease.
Signal · strong
Industry concentration
Most notable frontier models come from a handful of firms.
Signal · strong
AGI strategy divergence
Policymakers disagree whether AI is existential race or normal tech.
Signal · moderate
Compute as strategic asset
States treat clusters like dual-use infrastructure.
Signal · strong
What could happen next?
Scenario A
Managed diffusion
Standards and audits normalize powerful models as regulated utilities.
Scenario B
Race dynamics
Competitive pressure overrides caution; governance stays symbolic.
Scenario C
Fragmented sovereign stacks
Nations accept less capable local models for control.
Key companies / entities
- OpenAI
- Google DeepMind
- Anthropic
- US government
- China
Evidence
- research
AI Index concentration and compute findings
- policy
National AGI / frontier AI strategy papers
- news
Public risk polling and incident coverage
Prediction history
- Sep 22, 202652%
- Sep 15, 202650%
- Sep 8, 202647%
- Sep 1, 202645%