Open models aren't inherently safer—just 4-7 months behind closed ones, argues new post

BlackHC · x · 2026-11-15

Andreas Kirsch (BlackHC) publishes a systematic case for open models from a safety perspective.

Core claims

Open models are not inherently safer than closed ones—just roughly 4-7 months behind (6 months on cyber); wider distribution and thinner safeguards mean closed-model failure modes will show up in open models soon and more often.
Open weights add three risks: diffused responsibility, weaker lab safety commitments, and removable safety training.
Below dangerous-capability thresholds, open models are clearly net-beneficial: they capture most of AI's economic value, enable regional sovereignty and local defensive deployment, and spread gains widely.
Open models may slow the ASI race by diverting revenue and compressing margins at labs pursuing recursive self-improvement, though the author admits he's much less confident in this economic argument.

Governance proposal

The ideal weight-holder distribution is "peaky": a broad, modifiable open base below dangerous-capability thresholds, narrowing sharply into a thin spire of closely monitored actors above them.
The technical priority is safe-to-release open models: alignment that survives fine-tuning and abliteration, with dangerous capabilities absent and expensive to re-add.

He welcomes pushback, especially on the cruxes listed at the end.

More from AGI Musings