Near-frontier capability is open-weight and everywhere; no one can gate it.
The lag between gated frontier capability and open weights stays under a year. Any state, company or well-funded group can run a near-frontier model on its own hardware, fine-tune away safeguards, and deploy without oversight. Benefits diffuse widely (research, small business, non-English users), and so does misuse. Governance shifts from controlling models to defending targets.
Observable thresholds. When one crosses, the scenario's status changes and the weekly re-run is brought forward.
Every change is logged with its reason and the signals that drove it. Moves are bounded per week; a jump beyond the bound is flagged as a shock.
Disagree with our range? Set yours. Estimates feed a community view that appears once enough people weigh in, and the weekly run reads the gap between our number and yours.
Seed estimate, set to active as an ongoing process: GLM-5.3 (744B, MIT-licensed variant) shipped with strong cyber capability, Meta keeps an open agentic line, and Metaculus gives 95% that an open model matches the gated Mythos cyber benchmark by July 2027. Probability as the dominant trajectory is moderate because the very top stays closed.
Follows an August administration stance that frontier open models will need government collaboration, with an October deadline for standardized testing methods.
Meta continues an open-weight agentic line while shipping a stronger closed model.
Multiple vendors shipped small, fast or open-weight models in one week, continuing the diffusion of near-frontier capability at low cost.
Released after a two-week self-imposed cyber-safety hold. Vendor-reported 84.5% on CyberGym and thousands of real vulnerabilities found in open-source projects. Flash variant MIT-licensed.
Every source reviewed for this scenario. The full ledger is public.
| Date | Source | Publisher | Type |
|---|---|---|---|
| 2026-08-28 | Z.ai GLM-5.3 open weights (744B) released after cyber-safety hold | TechJack | news |
| 2026-09-09 | Forecasts after the Hugging Face incident, OpenAI breach, and cyberattack questions Open-weight matches Mythos cyber benchmark by Jul 2027: 95%. Another sandbox escape by Jan 2027: 50%. AI hacks third party: 41%. Weight exfiltration confirmed: 7%. Kill-switch bill passes both houses by Sept 2027: 24%. | Metaculus | forecast |
| 2026-08-10 | Introducing Muse Glimmer, an open agentic model | Meta | primary |
| 2026-09-08 | Intelligence agencies warn of China's large-scale AI model distillation efforts | Nextgov | news |
| 2026-09-15 | AI news feed (Bessent on liability; Gemini 3.8 Live; OpenRouter spend) | llm-stats | compilation |
Open models reach frontier offensive-cyber capability; a global surge in AI-driven intrusion follows.
Distillation and copying erase the frontier lead; ten-plus actors hold near-frontier capability.
An open model with removed safeguards materially helps someone cause a biological or chemical harm event.