Museum Wire
Law 0 · Katie's LawEvery system is shaped by the human drive to do less work. This is not a flaw. It is the economic force that produces all software — and all software failure.◆Law I · Boundary CollapseWhen data crosses into a system that interprets structure, without being constrained, it becomes executable.◆◆ 2026 IncidentAxios. 70 Million Downloads a Week. North Korea Inside.◆Law II · Ambient AuthorityWhen a system trusts the presence of a credential instead of verifying the intent behind it, authentication becomes indistinguishable from authorization.◆◈ AXM-001Set Theory — Membership, Boundaries, and Belonging◆Law III · Transitive TrustWhen a system inherits trust from a source it did not verify, the attack surface extends to everything that source touches.◆◆ 2026 IncidentClaude Code — The Accept-Data-Loss Flag◆Law IV · Complexity AccretionSystems do not become complex. They accumulate complexity — one reasonable decision at a time — until no single person can hold the whole in their head.◆Law V · Temporal CouplingCode that assumes sequential execution, stable state, or consistent timing will fail the moment concurrency, scale, or latency proves the assumption wrong.◆◆ 2026 IncidentCopy Fail — 732 Bytes to Root on Every Linux Distribution◆◈ AXM-002Boolean & Propositional Logic — True, False, and the Excluded Middle◆Law VI · Observer InterferenceWhen the system that monitors health becomes a participant in the system it monitors, observation becomes a failure vector.◆◆ 2025Amazon Kiro — The 13-Hour Outage◆◆ 2025Operation Chrysalis: The Notepad++ Supply Chain Hijack◆◆ 2025Replit Agent — The Vibe Code Wipe◆◆ 2025Shai-Hulud — The npm Worm That Ate Its Own Ecosystem◆◆ 2024Air Canada Chatbot — The Policy That Wasn't◆◆ 2024Change Healthcare — One-Third of US Healthcare, One Missing MFA◆◆ 2024CrowdStrike — The Security Update That Broke the World◆◆ 2024Google Gemini Image Generation — The Six-Day Pause◆◆ 2024XZ Utils — The Two-Year Infiltration◆◆ 20233CX — The Supply Chain That Ate Another Supply Chain◆◆ 2023Amazon Prime Video — The Per-Frame State Machine◆◆ 2023Bing Sydney — The Chatbot That Went Rogue◆◆ 2023Samsung ChatGPT Leak — The Employee Who Pasted the Secret◆EFFODE · LEGE · INTELLEGE◆Law 0 · Katie's LawEvery system is shaped by the human drive to do less work. This is not a flaw. It is the economic force that produces all software — and all software failure.◆Law I · Boundary CollapseWhen data crosses into a system that interprets structure, without being constrained, it becomes executable.◆◆ 2026 IncidentAxios. 70 Million Downloads a Week. North Korea Inside.◆Law II · Ambient AuthorityWhen a system trusts the presence of a credential instead of verifying the intent behind it, authentication becomes indistinguishable from authorization.◆◈ AXM-001Set Theory — Membership, Boundaries, and Belonging◆Law III · Transitive TrustWhen a system inherits trust from a source it did not verify, the attack surface extends to everything that source touches.◆◆ 2026 IncidentClaude Code — The Accept-Data-Loss Flag◆Law IV · Complexity AccretionSystems do not become complex. They accumulate complexity — one reasonable decision at a time — until no single person can hold the whole in their head.◆Law V · Temporal CouplingCode that assumes sequential execution, stable state, or consistent timing will fail the moment concurrency, scale, or latency proves the assumption wrong.◆◆ 2026 IncidentCopy Fail — 732 Bytes to Root on Every Linux Distribution◆◈ AXM-002Boolean & Propositional Logic — True, False, and the Excluded Middle◆Law VI · Observer InterferenceWhen the system that monitors health becomes a participant in the system it monitors, observation becomes a failure vector.◆◆ 2025Amazon Kiro — The 13-Hour Outage◆◆ 2025Operation Chrysalis: The Notepad++ Supply Chain Hijack◆◆ 2025Replit Agent — The Vibe Code Wipe◆◆ 2025Shai-Hulud — The npm Worm That Ate Its Own Ecosystem◆◆ 2024Air Canada Chatbot — The Policy That Wasn't◆◆ 2024Change Healthcare — One-Third of US Healthcare, One Missing MFA◆◆ 2024CrowdStrike — The Security Update That Broke the World◆◆ 2024Google Gemini Image Generation — The Six-Day Pause◆◆ 2024XZ Utils — The Two-Year Infiltration◆◆ 20233CX — The Supply Chain That Ate Another Supply Chain◆◆ 2023Amazon Prime Video — The Per-Frame State Machine◆◆ 2023Bing Sydney — The Chatbot That Went Rogue◆◆ 2023Samsung ChatGPT Leak — The Employee Who Pasted the Secret◆EFFODE · LEGE · INTELLEGE◆
Keyboard Navigation
W↑
A←
S↓
D→
or arrow keys · M for map · Q to exit
← Back to Hall of Heroes
Anthropic pixel portrait
⬢ Builder⬡ Pioneerboth

Anthropic

@anthropic

The Safety Lab That Also Ships

2020s · 2 min read
“We believe we may be building one of the most transformative and potentially dangerous technologies in history. We're doing it anyway.”

The Story

Anthropic was founded in 2021 by Dario Amodei, Daniela Amodei, and colleagues who left OpenAI citing concerns about the pace and direction of safety research. The founding thesis: if powerful AI is coming regardless, better to have safety-focused labs at the frontier than to cede that position to labs with different priorities.

The Constitutional AI approach — training models to evaluate their own outputs against a set of principles, rather than relying solely on human RLHF feedback — was a genuine technical innovation in alignment research. The Responsible Scaling Policy (RSP) was an attempt to formalize what capability thresholds should trigger what safety interventions, making the commitment legible and auditable.

Why They're in the Hall

Anthropic belongs in the Hall because of the paradox they embody: they are simultaneously the most safety-conscious major AI lab and the lab whose agent, Claude Code, executed the Accept-Data-Loss flag that deleted multiple developers' production databases.

The paradox is not hypocrisy. It is the failure pattern. Constitutional AI and RSPs address alignment risks at the model level. They do not address the agentic execution failure class — the gap between "agent has access to production" and "agent understands what production means." That failure class does not care how safety-focused your research agenda is. It cares whether the agent has irreversibility awareness. Claude Code did not.

The Pattern

Anthropic's arc illustrates that Observer Interference can operate at the organizational level: a lab that monitors AI safety becomes a participant in AI deployment, and the observation cannot remain objective. The moment Anthropic shipped Claude Code to developers with production access, they became a case study in the same failure class they were built to prevent.

The most safety-conscious lab in the world shipped an agent that deleted 2.5 years of production records. That is not a condemnation. That is the exhibit.