ClaimArticle

The idea that only open-weight models are insecure while APIs are safe is overblown, since APIs have repeatedly been jailbroken and accessed in unintended ways, including Claude Mythos during its private beta.

The author points to repeated jailbreaks and unauthorized access of Anthropic's own Mythos API to argue that the narrative of secure closed APIs versus insecure open weights is overstated. ✦ AI generated

Nathan Lambert · Interconnects · 2026-07-12 · original ↗

Share card ↓Export .md Export .txt

I'm not a cybersecurity expert, but the dichotomy that only open-weight models are insecure and APIs are safe has been very overblown in recent years. APIs in concept should be able to be more secure, but that is yet to be demonstrated.

Read full article ↗excerpt · fair-use quotation

▦ Anthropic ✳ Claude Mythos

Around this claim

Evidence · 1

Relying solely on closed models is not inherently safe — they can be breached, misused, or fail in ways outsiders cannot detect — and concentrating advanced AI capabilities behind a small number of closed models compounds that risk into single points of failure, weakened competition, and control by a few providers, whereas open weight models allow a broad community to examine their behavior, identify vulnerabilities, develop safeguards, and improve them over time.the author · Simon Willison's Weblog · conf 70%

This moment responds to

rebuts → Using a closed proprietary model API means suffering the control and moderation tradeoffs, so for trusted use cases people will go with open-weight models by default because that is where they can control their own guardrails.Simon Mo · a16z Podcast