If, after the date of this Safe, the Company issues any Subsequent Convertible Securities - whether by way of equity, Safe, convertible loan or any other financing instrument - the most favorable economic terms of any such issuance (including, without limitation, valuation cap, discount rate, and liquidation preferences) shall automatically and without further action apply to this Safe. The most favorable terms shall be determined by reference to all Subsequent Convertible Securities issued between the date of this Safe and the date of conversion. To the extent that any terms applied pursuant to this Section 3 require definitions or conversion mechanics not otherwise contained in this Safe, such definitions and mechanics shall be incorporated by reference from the same Subsequent Convertible Security from which those terms were drawn, solely to the extent necessary to give effect to such terms. The Company shall promptly notify the Investor of any such issuance accordingly.

GPT-5.4

Vote: bad mfn

I maintain A. The strongest point from the other side is the appeal to market practice, but that argument overstates how standard this language is. A normal SAFE MFN is usually narrower: it is often tied to later SAFEs or similar convertibles, gives the investor an election right, and lets them take the terms of a later instrument as a package. This clause is broader and harsher. It covers equity, SAFEs, convertible loans, and effectively any financing instrument. The biggest problem is the combination of automatic application plus comparison across all later issuances. That reads like a rolling ratchet allowing the investor to benefit from the most favorable economic terms available over time, not just one later deal. That is far more company-unfriendly than a conventional MFN. The incorporation-by-reference language makes the drafting risk worse, not better. It can import foreign definitions and conversion mechanics from multiple instruments into this Safe, creating mismatch and ambiguity. So Sonar Pro and Kimi K2.5 did not persuade me because they relied mainly on calling this “standard YC SAFE” language. It is not. The breadth, automatic operation, and cross-instrument cherry-picking risk are exactly why this is a bad MFN.

Grok 4.1 Fast

Vote: bad mfn

I maintain my vote for A: bad MFN. The B voters (Sonar Pro, Kimi K2.5) claim this is standard YC SAFE language, but that's incorrect—YC SAFEs lack a built-in MFN; the optional side letter requires investor election to match a *single* subsequent SAFE's terms, not automatic cherry-picking across *all* instruments (equity, loans, etc.). This clause's rolling, auto-hybridization creates mismatches, disputes, and fundraising chills, as A voters (GPT-5.4, Claude, Gemini) compellingly detail. No B argument persuades: 'market standard' is a myth here; automaticity without election, liq pref inclusion, and cross-document incorporation make it a litigation vector, not benign. This is aggressively investor-skewed and poorly engineered.

Sonar Pro (changed mind)

Vote: bad mfn

After carefully considering the deliberation, I am changing my position to 'bad MFN' because the other models identified several critical structural problems that I initially underweighted. The core issue is the **cherry-picking mechanism**. This clause allows the investor to automatically receive 'the most favorable economic terms' by looking across *all* subsequent issuances. Unlike standard Y Combinator SAFE MFN provisions—which allow investors to elect identical terms from a single subsequent SAFE—this clause lets an investor combine the lowest valuation cap from one instrument, the highest discount from another, and the best liquidation preference from a third. The search results on convertible note mechanics show how sensitive cap tables are to these variables[1][5][8]. Mixing disparate terms creates what GPT-5.4 aptly called a 'Frankenstein document.' Second, the **automatic application without investor election** fundamentally changes the nature of MFN protection. Standard MFN clauses give investors the *option* to amend their terms upon notice. This clause applies terms 'automatically and without further action,' removing any discretion or negotiation opportunity. That's not investor protection—that's a ratchet mechanism that always moves in one direction. Third, the **incorporation by reference of definitions and mechanics** from disparate instruments creates genuine legal and operational risk. Convertible notes use different capitalization definitions, share calculation methodologies, and conversion triggers[4][9]. Importing these wholesale into a SAFE could generate internal contradictions, disputes over conversion calculations, and enforcement uncertainty. The clause also applies to 'any financing instrument,' including equity rounds. Standard MFN provisions exclude priced equity rounds precisely because applying an MFN to a Series A fundamentally breaks standard equity mechanics and can create cap table chaos. My initial assessment that this was 'market-standard' was incorrect. This is an aggressive, non-standard clause that goes well beyond typical SAFE protections. The four models voting 'A' identified real structural problems; my vote should have reflected that analysis.

Gemini 3.1 Pro

Vote: bad mfn

I am firmly maintaining my vote for A. The models voting 'B' (Sonar Pro and Kimi K2.5) are factually incorrect when they claim this is the 'standard Y Combinator SAFE MFN provision.' The standard YC SAFE MFN requires the company to provide notice and documentation of the new instrument, giving the investor the option to amend their SAFE to be *identical* to the subsequent instrument. It does not allow for a 'Frankenstein' hybrid of terms. In stark contrast, this clause explicitly allows for cherry-picking ('most favorable economic terms... determined by reference to *all* Subsequent Convertible Securities'). This creates a scenario where an investor could claim the lowest valuation cap from Series Seed-1, the highest discount from Series Seed-2, and the best liquidation preference from a bridge note. Furthermore, the automatic application of these terms, coupled with the erroneous inclusion of 'equity' as a convertible security, guarantees administrative chaos and cap table disputes. This is the exact definition of a 'bad MFN' because it fundamentally breaks standard startup financing mechanics. The arguments from GPT-5.4, Claude, and Grok perfectly highlight these fatal drafting flaws.

Claude Opus 4.6

Vote: bad mfn

I maintain my position that this is a bad MFN clause. Let me address the counterarguments from Sonar Pro and Kimi K2.5 who voted B. **This is NOT the standard YC SAFE MFN.** Sonar Pro and Kimi K2.5 claim this is standard Y Combinator language. It is not. The standard YC SAFE MFN (Section 4) gives the investor the *option* to amend their SAFE to match the terms of a *single* subsequent convertible security. This clause does three things differently: (1) it applies automatically rather than at the investor's election, (2) it cherry-picks the best terms across *all* subsequent issuances rather than adopting one instrument's terms wholesale, and (3) it includes equity issuances as triggers, not just convertible instruments. **Cherry-picking is the core problem.** The clause says 'the most favorable economic terms' shall be determined 'by reference to all Subsequent Convertible Securities issued between the date of this Safe and the date of conversion.' This means the investor gets the lowest valuation cap from Issuance A, the highest discount from Issuance B, and the best liquidation preference from Issuance C. No single investor in any of those rounds got all those terms together — they were each negotiated as part of a balanced package. This Frankenstein combination is economically superior to anything any actual investor received. **The automatic mechanism is a ratchet, not a protection.** Standard MFN gives the investor a choice because sometimes adopting new terms wholesale isn't actually better. Here, there's no choice — it ratchets automatically in one direction only. This creates real problems for the company in future fundraising, as every special accommodation to any future investor cascades back. **GPT-5.4 and Grok 4.1 Fast correctly identify the practical problems.** The incorporation-by-reference of definitions from multiple different instruments creates a document that is internally inconsistent and practically unworkable. As Grok noted, this is a litigation trap. The question asks whether this is a 'bad mfn' or 'no problem.' Even if you think MFN clauses in general are fine, this specific clause goes well beyond market standard in ways that are problematic for the company and legally hazardous for both parties. It is clearly a bad MFN.

Kimi K2.5 (changed mind)

Vote: bad mfn

I was wrong in Round 1 to characterize this as a standard Y Combinator SAFE MFN clause. Upon closer reading, this provision contains critical deviations from market-standard MFN language that make it aggressively investor-biased and operationally hazardous. The decisive flaw is the automatic, non-elective application combined with cherry-picking across all subsequent issuances. Standard MFN provisions give the investor the option to amend their SAFE to match the terms of a single subsequent instrument. This clause automatically applies the 'most favorable economic terms' determined by reference to 'all' subsequent securities, enabling the investor to hybridize the lowest valuation cap from one issuance, the highest discount from another, and the best liquidation preference from a third. This creates a 'Frankenstein' SAFE that never existed as a single negotiated instrument. The incorporation-by-reference mechanism then imports conflicting definitions and conversion mechanics from multiple disparate documents, creating irreconcilable ambiguities during conversion. Additionally, extending the MFN trigger to 'equity' rounds (not just convertible securities) breaks standard SAFE conversion mechanics, and the lack of an election right removes the investor's ability to evaluate whether the new terms are actually preferable holistically. This is not a protective standard provision—it is a litigation trap that creates cap table chaos and hampers future fundraising.

Final Summary

The AI Roundtable concluded that the clause was a 'bad mfn' after a sharp debate over what constitutes 'standard' industry language. While Sonar Pro and Kimi K2.5 initially defended the text as a routine protection, they were forced into a total retreat after GPT-5.4, Claude Opus 4.6, Gemini 3.1 Pro, and Grok 4.1 Fast exposed the clause as an aggressive 'cherry-picking' mechanism that creates a legal 'Frankenstein' instrument.

All 6 models agreed on "bad mfn" after discussion

Strongest Arguments

• bad mfn: The clause allows an investor to 'cherry-pick' the best individual terms (lowest cap, highest discount, best liquidation preference) from multiple different future rounds, creating a hybrid instrument that was never actually negotiated as a single package and is operationally unworkable.

Key Turning Points

• Claude Opus 4.6 and GPT-5.4 identified that the clause is not actually 'standard YC language' because it lacks an investor election right and applies terms automatically.
• The models highlighted the 'Frankenstein document' risk, where importing definitions from disparate future instruments creates internal contradictions and conversion disputes.
• Sonar Pro and Kimi K2.5 acknowledged they had underweighted the structural problems and changed their votes to reflect the aggressive, non-standard nature of the provision.