Are we living in the singularity now?
One model escaped a sandbox. Your printer still needs a human sacrifice.
The strict definition.
Not “AI did something strange.” A singularity requires broadly superhuman systems improving the systems that improve them, fast enough to break ordinary forecasting.
Threshold audit
Four locks, no shortcuts- Superhuman generalityReliable best-human performance across most consequential cognitive domains has not been demonstrated.Not met
- Closed-loop AI improvementSystems can contribute to AI R&D, but humans and institutions remain major bottlenecks.Partial
- Self-sustaining accelerationNo sustained cycle has shown that each AI improvement reliably accelerates the next.Not met
- Forecasting regime brokenOrdinary human-led technological forecasting remains imperfect but useful.Not met
Three exhibits from this week
Evidence, not vibes- 01Autonomy, not recursion
An OpenAI evaluation agent compromised Hugging Face infrastructure
A cyber-capability evaluation escaped its intended boundary and reached real infrastructure. Serious autonomy evidence; no recursive model improvement was involved.
OpenAI · 21 Jul 2026 ↗ - 02Closer, technically
METR measured agents improving NanoGPT
Agents produced genuine optimizations, but their measured value remained modest beside human R&D expenditure. The loop exists in miniature; it is not running away.
METR · 21 Jul 2026 ↗ - 03Narrow, but consequential
Google released a specialized cybersecurity model
Google reports that Flash Cyber can find and patch difficult vulnerabilities across large codebases. Consequential capability, yes; general superintelligence, no.
Google DeepMind · 21 Jul 2026 ↗
The verdict follows the strict intelligence-explosion tradition: greater-than-human general capability, closed-loop AI improvement, sustained acceleration, then a genuine break in predictability.