Home › Q&A › My product runs on a single frontier API

My product runs on a single frontier API: what if it gets cut off? Do I need a second supplier now?

Asked by:创始人·依赖 Claude / GPT 的应用 CTO · Judged · 张倩 Cynthia Zhang · FutureX Capital

Yes, schedule it this quarter: a dual-supplier abstraction layer, swappable weights, and a fallback path for critical capabilities. A cutoff has already happened once: a US Commerce Department letter on 6/12 put Anthropic's Fable 5 under control, and it returned globally on 7/1, a full 18 days offline.

Judgment

The mainstream view treats a cutoff as a technical risk. The diary books it as administrative-discretion risk: your access sits in another country's hands and one letter can end it. Current status first: ENTRY 85's "tiered by nationality" call was corrected by ENTRY 91 as only half right. The wall is built on identity, and the restoration was global.

Facts

The US Commerce Department's 6/12 letter cited EAR §744.22(b) and required export licenses for Anthropic's Fable 5 and Mythos. Commerce lifted the export control on 6/30 and Fable 5 returned globally on 7/1, a full 18 days offline; from 7/8 the full-capability consumer version requires identity verification, while API users are exempt. Caveat: the statutory export license fell case by case on Anthropic alone; OpenAI and Google cooperated voluntarily.

What to do

For founders: schedule the dual-supplier abstraction layer and fallback path this quarter, keep core business on the API side, and leave a slot for identity verification in consumer products. For investors: in diligence, ask how long it takes to switch away if the main model is cut off, and discount as a single point of failure if there is no answer; use 18 days as the pricing anchor instead of cutting valuation as if the damage were permanent.

Falsifier (verbatim from the diary footnote)

Only if the capability threshold is written into a formal ECCN rule within one year and applied to frontier models generally does "case by case" count as "a regime".
Ledger status:ENTRY 108 台账 ⏳ 进行中。ENTRY 85 台账 ⏳ 进行中(部分验证+部分修正):「按国籍分级」半句已由 ENTRY 91 公开更正为「身份墙」。ENTRY 91 未入台账,其尾注回滚点(非证伪口段):起算日 2026-07-08,若验证上线后按国籍收紧,「身份墙」升级回「国籍墙」,以 Anthropic 官方验证条款为裁定源,这笔账重算。
Disclosure: This diary is written and verified with Claude; discount any judgment about Anthropic, favourable or not; FutureX Capital does not, on publicly available information, hold the companies discussed here.
Source entries:ENTRY 108 · 前沿 AI 被正式当军火管了:同一个底座,一个加护栏卖给大众,一个开满能力只给批准机构——最强的智能不再有公开版本 · ENTRY 85 · 7 月 8 日起交护照和脸——前沿 AI 从「付费墙」走向「护照墙」 · ENTRY 91 · 被拔线 18 天后,Fable 5 回来了——赌盘把日期猜对了,我把墙的材质猜错了一半
Related questions:
Topic pages:Anthropic · Claude · US–China AI policy · export controls · OpenAI · GPT