# EIP-8: Invite Liability for AI Slop / Bot Accounts

- Space: [Ethos](https://snapshot.box/s:ethos-labs.eth.md)
- Author: 0xe505570cCa5734c3D803a9cFE0FBCE618A0d74D1
- State: Passed
- Voting type: Basic
- Start: 2025-08-05 22:59 UTC
- End: 2025-08-08 22:59 UTC
- Votes: 315
- Open in Snapshot: https://snapshot.box/#/s:ethos-labs.eth/proposal/0xda6239ea33ca41ac4b6884b7a2ae89051172639e869b35bbbdb87ba8b517a43f

## Results

- For: 313 voting power, 99.4%
- Against: 0 voting power, 0.0%
- Abstain: 2 voting power, 0.6%

## Proposal

![6919991a-d8d2-4713-9f10-c43c741fc938-1.jpg](ipfs://bafybeihfwgcvh23onp5qomqpipoudx2xdxu6pcvuitn2dahw3jfxw6nk5e)
## Summary
Introduces a penalty + safeguard system for inviters who onboard AI or bot-controlled accounts. It adds multi-sig bot detection, a 7-day remediation window, a 23-day public appeal process, escalating strikes on negligent inviters, and an Invite-Bot Ratio lock.  It shifts bot-fighting one layer earlier, discouraging bad actors before they gain traction while protecting legitimate users from false blame.
---

## 1 Motivation
* **Bots dilute credibility** - AI-generated "slop" reviews and automation cheapen Ethos signals.  
* **Invite is the Sybil throttle** - The 90-day, 20% score bond already puts skin in the game for inviters, we extend it with explicit bot penalties.  
* **Consistent progression** - Previous EIPs neutralize content after posting, this EIP targets bad actors before they get traction.  

---

## 2 Definitions & Context
| Term | Meaning |
|------|---------|
| **Inviter** | Ethos profile that sent the on-chain invite. |
| **Invitee** | Address that accepted the invite. |
| **AI-Bot** | Invitee meeting ≥ 2 detection signals (§3.1) *and* losing its appeal. |
| **Strike** | Recorded against the inviter when an invitee is confirmed AI-Bot. |
| **IBR (Invite-Bot Ratio)** | `Confirmed Strikes ÷ Accepted Invites` (rolling 12 months). |
| **EIP-7** | Sets review impact to 0 for "spam / review-for-review" text. |
| **EIP-6** | Deducts points when recipients down-vote bogus positive reviews. |

---

## 3 Specification

### 3.1 Detection Signals – any **two** trigger §3.2
1. ≥ 70% of invitee reviews flagged "Spam" (EIP-7).  
2. ≥ 3 positive reviews down-voted by recipients within 30 days (EIP-6).  
3. No attestations, vouches or wallet signatures after 30 days.  
4. Non-human behaviour pattern (24×7 posting, low perplexity, copy-paste).

### 3.2 Timeline & Appeal
| Day / Window | State | Action |
|--------------|-------|--------|
| **0** | `Suspected Bot` | Detection engine flags account → status **Suspected Bot** (no penalty yet). |
| **0-7** | Quarantine | Invitation link is frozen. Inviter and invitee receive on-chain alert. Invitee can clear the flag by passing **CAPTCHA + wallet-signature**. |
| **7** | `Confirmed Bot` | If still flagged: invitee score → 0 and a **tentative Strike** is applied to the inviter. The account now becomes eligible for appeal. |
| **7-30** | Appeal Window | At any time in this window the invitee may file a public **Plea** (UI mirrors Slash). Filing the plea triggers a **72-hour community vote**. |
| **Plea + 3 days** | Voting Closes | After exactly 72 hours the vote is tallied. Outcome determines whether the flag and Strike are removed or upheld. |
| **No Plea by Day 30** **or** **Vote Result** | Final | *No Plea*: Strike auto-locks on day 30. *With Plea*: once the 72-hour vote ends, the decision is final. No further appeals. |


### 3.3 Penalty Ladder (for the **inviter**)
| Strike # | Score Loss | Invite Freeze |
|----------|------------|---------------|
| 1 | −25 pts | — |
| 2 | −50 pts | 7 days |
| 3 | −100 pts | 30 days |
| 4+ | Fibonacci (0,1,1,2,3…) × 50 pts (cap −400) | 90 days plus DAO review |

*Strike is tentative at Day 7 and becomes permanent if no appeal is filed by Day 30 or if the 72-hour community vote upholds the bot flag.*

### 3.4 IBR Limit (plain English)
If more than **15%** of your accepted invites become confirmed bots in the last 12 months, your invite button locks until the ratio drops below 15%.  
*Example: 40 invites allow up to 6 bots, a 7th freezes inviting.*

---

## 4 Safeguards
* Two-signal rule cuts false positives.  
* 7-day remediation lets real humans prove themselves quickly.  
* Public vote aligns with existing governance, transparent and inexpensive.  
* Strikes decay after 12 months, rewarding improved diligence.  

---

## 5 Future Considerations
* **Proof-of-Personhood (KYC-lite, opt in):** verified humans earn a small permanent score bonus.  
* Dynamic ML thresholds to catch new AI patterns without further EIPs.  
* Light liability for early *vouchers* of bots after 90 days.

---

To vote or create proposals from an AI assistant, connect the Snapshot MCP at https://mcp.snapshot.box

