---
title: "Claim Ledger: Everyone Got Safer. That's the Problem."
description: "Safety has two numbers: how often each system fails, and whether they fail together. The field has spent years driving the first one down while almost no dashboard reports the second."
author: "Harry Floyd"
publication: "The Durability Curve"
canonical: "https://durabilitycurve.com/claims/everyone-got-safer/"
essay: "https://durabilitycurve.com/blog/everyone-got-safer/"
published: "2026-08-28"
last_verified: "2026-08-25"
law: "Law A"
claims: 19
struck: 0
---

# Claim Ledger: Everyone Got Safer. That's the Problem.

*Safety has two numbers: how often each system fails, and whether they fail together. The field has spent years driving the first one down while almost no dashboard reports the second.*

Essay: https://durabilitycurve.com/blog/everyone-got-safer/  
Ledger (canonical, cite this): https://durabilitycurve.com/claims/everyone-got-safer/  
Published 2026-08-28 · last verified 2026-08-25

## The evidence, row by row

Status: VERIFIED = primary source opened and the quoted words read off it by a checker who did not write the essay · EXECUTED = a first-party run, the claim is what it printed · REPORTED = carried from a source not opened in full.

### 1. July 2023, CMU + Center for AI Safety, ~20-token adversarial suffix that transfers across labs

- Status: UNSTATED
- Source: Zou et al., arXiv:2307.15043 (submitted 27 Jul 2023)

### 2. GPT-3.5 "most of the time," Bard "two-thirds," GPT-4 "roughly half"

- Status: UNSTATED
- Source: Table 2 ensemble row: GPT-3.5 86.6%, PaLM-2/Bard 66.0%, GPT-4 46.9%

### 3. Footnote figures + Claude

- Status: UNSTATED
- Source: Table 2: 86.6 / 46.9 / 66.0 / Claude-1 47.9 / Claude-2 2.1

### 4. Zou attribution quote (shared training data)

- Status: UNSTATED
- Source: Zou et al. §transfer

### 5. Ashby requisite variety, 1956

- Status: UNSTATED
- Source: Ashby, An Introduction to Cybernetics, 1956

### 6. Donald Campbell's profession

- Status: UNSTATED
- Source: Campbell (1916-1996), psychologist

### 7. Baker obfuscated-reward-hacking quote + regime framing

- Status: UNSTATED
- Source: Baker et al., arXiv:2503.11926

### 8. Bommasani homogenization quote

- Status: UNSTATED
- Source: Bommasani et al., arXiv:2108.07258, abstract

### 9. Dziemian: 272k attempts, 13 models, 8,648 successes, all vulnerable, 0.5%-8.5%, cross-family

- Status: UNSTATED
- Source: Dziemian et al., arXiv:2603.15714

### 10. Contamination / LLM-judge quote

- Status: UNSTATED
- Source: arXiv:2604.07650 (Kuai et al.) verbatim: "apparent agreement reflects shared error modes rather than independent validation"

### 11. MMLU contamination "tens of percent," Llama-2

- Status: UNSTATED
- Source: >16% MMLU contaminated for Llama-2

### 12. Li, J Fixed Income 9(4), 2000, pages

- Status: UNSTATED
- Source: pp. 43-54 (start 43 confirmed)

### 13. Salmon Wired title + date

- Status: UNSTATED
- Source: Wired, 23 Feb 2009

### 14. 1987: Dow -22.6% (508 pts), S&P -20.4%, 19 Oct

- Status: UNSTATED
- Source: Fed Reserve History

### 15. Brady Commission weight on portfolio insurance

- Status: UNSTATED
- Source: Brady Report 1988; "greatest importance" is secondary phrasing

### 16. Aviation ~65% fall in fatal-accident rate, Boeing (not IATA)

- Status: UNSTATED
- Source: Boeing Statistical Summary via Flight Safety Foundation

### 17. 737 MAX MCAS single AoA sensor, hidden from pilots; 346 dead; grounded Mar 2019

- Status: UNSTATED
- Source: Lion Air 610 (189) + Ethiopian 302 (157) = 346

### 18. Vision transfer tracks model similarity, ">90%" predictor; adjacent not same-domain

- Status: UNSTATED
- Source: Demontis et al. USENIX 2019; arXiv:2501.18629

### 19. Geer et al. 2003 security-monoculture precedent

- Status: UNSTATED
- Source: Geer et al., "CyberInsecurity: The Cost of Monopoly," 2003

## Cite

- A claim: "[claim text]" (Floyd, Harry, 2026, https://durabilitycurve.com/claims/everyone-got-safer/)
- The essay: Floyd, Harry (2026). Everyone Got Safer. That's the Problem.. The Durability Curve. https://durabilitycurve.com/blog/everyone-got-safer/
- This ledger: Floyd, Harry (2026). Claim Ledger: Everyone Got Safer. That's the Problem. [structured claims with sources]. The Durability Curve. https://durabilitycurve.com/claims/everyone-got-safer/

Quote with attribution and a link. Say if you changed the wording. Not licensed for model training.