What Is Anthropic LTBT? The Long-Term Benefit Trust and Governance Boundaries

What is LTBT? Read this short answer first

LTBT stands for Long-Term Benefit Trust, which can be called the long-term benefit trust in Chinese; it is part of Anthropic’s corporate governance arrangements. It serves different purposes from the Constitution used to train Claude and the user-facing product policies, and the three cannot be treated as the same set of rules.

The 2023 establishment explanation explained the design of the special class of shares, Class T, and the authority to appoint and remove directors;the official announcement dated October 21, 2026 expressly disclosed that the Trust has the authority to appoint a majority of directors. The current company page lists the board of directors and the Trust’s trustees separately, and the two should likewise not be conflated.

Here it is necessary to distinguish institutional authority from actual exercise: having appointment authority does not mean that this authority has already been exercised in a particular model release, commercial transaction, or day-to-day operation. Governance effectiveness cannot be inferred solely from the company’s mission statement, nor should the old draft’s summary of “overriding shareholders” be understood as unlimited control over all corporate matters.

Read directly by question

If you want to see the original design and Class T:jump to the original LTBT section, then compare it with the official materials above.
If you want to check the current list of trustees and directors: look at the company page, verify the two lists separately, and record the verification date.
If you want to assess whether governance is effective: evidence of specific appointments, decisions, and accountability is also required; an institutional introduction cannot replace this step.

This section was verified on 2026-10-25. The chronological order also needs correction: LTBT was publicly announced in September 2023, before OpenAI’s board event in November of the same year, so the causal claim in the historical compilation below that “LTBT was designed after learning from that event” does not hold. The original peer comparisons, valuations, and effectiveness judgments are preserved as the historical compiler’s views; they were not updated item by item this time and cannot serve as a current corporate governance ranking.

Original compiled text and sources

When to open: You want to understand how Anthropic turns “mission / safety / governance” into verifiable institutions, not promotional rhetoric. This wiki covers 4 specific mechanisms + 1 public-interest project + 1 organization page.

One-sentence core point:Anthropic’s “atypical AI company” characteristic is not merely stated; it is implemented through 4 institutions: —— ① Claude Constitution(model training principles made public) ② Transparency Hub(data / policy / algorithm disclosure) ③ LTBT Long-Term Benefit Trust (structurally placing the mission above shareholders in governance) ④ Economic Futures Programme(connecting research and policy with the UK + Europe). These are externally verifiable institutions, not PR talking points.

Boundaries with other wikis: Constitutional AI (training method) + Core Views + RSP / ASL + Welfare + Red Team are in Anthropic AI Safety / Alignment / Policy Research Complete Collection (Constitutional AI + RSP + Welfare + Policy); the 5 major Economic Index data report series are in Anthropic Economic Index Complete Collection (AI Economic Impact Measurement, 2025–2026 Series); enterprise products + partnerships are in Anthropic Enterprise / Industry Product Portfolio + Strategic Partnerships (2025–2026). This wiki focuses on institutions / governance / policy.

0. Overview of 9 sources

Source	Type
`policy/constitution/`	Claude Constitution(Claude training principles)
`policy/transparency/`	Transparency Hub(unified disclosure portal)
`news/the-long-term-benefit-trust/`	LTBT governance structure (established in 2023-09)
`economic-futures/`	Economic Futures landing page
`economic-futures/program/`	Economic Futures Project structure
`economic-futures/symposium-proposals/`	Economic Futures Academic symposium proposal application
`news/economic-futures-uk-europe/`	UK + Europe launch announcement
`careers/`	Careers landing
`company/`	Company landing

1. Claude Constitution (model training principles made public)

From policy/constitution/.

1.1 What is the Claude Constitution

Claude Constitution = Anthropic The publicly available set of principles used in training Claude.

Relationship with the Constitutional AI training method(see Anthropic AI Safety / Alignment / Policy Research Complete Collection (Constitutional AI + RSP + Welfare + Policy) §1):

Constitutional AI method: an algorithm for training a harmless model using a set of principles
Claude Constitution:The specific content of that set of principles, publicly readable

1.2 Why make it public?

Anthropic’s logic:

Opaque training rules → users cannot determine the source of model behavior
Transparent training rules → users can understand / criticize / improve
This follows the same logic as Google Search Quality Rater Guidelines(public since 2015) — making the rules seen by raters public and accepting external scrutiny

1.3 Cross-check with the Karpathy LLM Wiki Pattern

Karpathy emphasizes “build AI systems where prompts and rules are inspectable code, not opaque magic” (see Karpathy System (Cross-Author Thematic Synthesis)).Claude Constitution is precisely this kind of inspectable rules.

The detailed principles are in the source files; I do not repeat them in the wiki (to avoid staleness).

2. Transparency Hub (unified disclosure portal)

From policy/transparency/.

2.1 What is the Transparency Hub

Transparency Hub = Anthropic Centralized disclosure of:

model training data policies
algorithmic decision-making impact
government request statistics (similar to the Google Transparency Report)
ASL level trigger records
links to major incident postmortems (see Anthropic Engineering Team First-Party Perspective: Agent Engineering / Skills / MCP / Claude Code (2025–2026 Complete Collection) §6.2 The two Claude Code postmortems)

2.2 Relationship with the Detecting Misuse quarterly reports

Anthropic AI Safety / Alignment / Policy Research Complete Collection (Constitutional AI + RSP + Welfare + Policy) §3.1’s “quarterly misuse report” is a subproject. The Hub is the unified entry point; detailed incident reports are in the dedicated project.

2.3 Industry comparison

Company	Transparency hub
Anthropic	Transparency Hub + quarterly misuse report + 2 Claude Code postmortems + Core Views (see Anthropic AI Safety / Alignment / Policy Research Complete Collection (Constitutional AI + RSP + Welfare + Policy) §2.1)
OpenAI	Usage Policies + System Card (released for each model generation) + Preparedness Framework
Google	Transparency Report (mainly Search / YouTube content removals)
Meta	Community Standards Enforcement Report

Anthropic’s distinctive feature:brings all three—model behavior + training principles + incidents—into transparency; most of the industry only makes the content moderation portion transparent.

3. The Long-Term Benefit Trust(LTBT,2023-09)

From news/the-long-term-benefit-trust/.

3.1 What is LTBT

LTBT(Long-Term Benefit Trust) = the governance structure, publicly announced in 2023-09.

Core question (as stated by Anthropic): Transformative AI will bring unique challenges + long-term opportunities, and traditional corporate governance structures (accountability to shareholders) are insufficient. It needs to be fine-tuned so that the long-term mission is not overwhelmed by short-term commercial pressure.

3.2 How LTBT works

Mechanism(simplified version; see the original text for details):

Establish independent Trust(trust)
The Trust holds / controls Anthropic’s special voting shares(Class T)
Class T shares have appoint / remove directors, above ordinary investors
The Trust is composed of independent trustees (not accountable to Anthropic investors / employees, but to the mission)

3.3 Comparison with Other AI Companies’ Governance Structures

Company	Governance structure
Anthropic	LTBT (independent Trust + Class T overriding shareholders)
OpenAI	"Non-profit controlling stake + capped-profit subsidiary," after the November 2023 crisis Sam Altman was briefly removed by the board and then reinstated, exposing governance fragility
DeepMind	Wholly owned Google subsidiary, with no independent governance
xAI	Dominated personally by Musk

Anthropic’s logic:The OpenAI November 2023 crisis proved that the "nonprofit parent company" governance design was not robust enough — LTBT is the "more robust" design learned from that.

3.4 Industry skepticism (explicitly listed)

"The Trust itself could also be corrupted": trustees are people and could also be bought off / replaced
"Commercial pressure will ultimately win": after Anthropic’s $350 billion valuation, can LTBT still hold back commercial interests?
"Blurred boundaries of independence": How does the Trust select trustees? How is it accountable?

This wiki does not make a judgment; it explicitly lists the disagreements.

4. Anthropic Economic Futures Programme (policy program)

From economic-futures/ + program/ + symposium-proposals/ + news/economic-futures-uk-europe/.

4.1 The program itself

Anthropic Economic Futures Programme — turning Economic Index data into a multinational policy program.

Structure:

Funding UK + European academia / think tanks to conduct research on AI’s economic impact
Sharing Anthropic data + tools(in part, under a compliance framework)
Engaging with policymakers(UK / EU)
Academic symposium: accepting proposals and organizing events

For detailed economic data, see Complete Anthropic Economic Index Collection (Measuring AI’s Economic Impact, 2025–2026 Series) §4 + the five major reports.

4.2 program / application

economic-futures/program/: detailed program information.

economic-futures/symposium-proposals/:Academic symposium proposal application page — scholars can apply to host a symposium on AI economics topics, funded by Anthropic.

4.3 Significance

Not just publishing papers, but turning research into an ongoing policy mechanism
Covering the UK + Europe (policy-setting direction), cross-checked with Complete Collection of Anthropic AI Safety / Alignment / Policy Research (Constitutional AI + RSP + Welfare + Policy) §7 "2028 Two Scenarios" policy positions
Along with the $200M collaboration with the Gates Foundation (global health / education / agriculture; see Anthropic Enterprise / Industry Product Portfolio + Strategic Partnerships (2025–2026) §2.2), this forms a dual-track "public good + policy" influence model

5. Careers + Company landing

5.1 Careers

careers/ — Anthropic recruitment portal. What this batch of static crawling obtained was the landing framework (Greenhouse embed, job list loaded by JS,the specific job listings were not captured).

Significance: Anthropic’s hiring pace and main roles (research / engineering / policy / safeguards) reflect the company’s strategic priorities.They can only be observed indirectly through other channels (events / LinkedIn / X posts).

5.2 Company

company/ — Anthropic company landing page. Includes the mission statement, leadership (Dario / Daniela Amodei, etc.), and office locations.

Significance:A unified external image and disclosure entry point; together with LTBT + Constitution + Transparency Hub, it forms "four layers of external disclosure" (governance / training principles / operational transparency / basic company information).

6. Anthropic Governance + Transparency vs. OpenAI Governance (Comparison After the November 2023 Crisis)

Dimension	Anthropic	OpenAI
Governance structure	LTBT independent Trust + Class T shares	"Non-profit controlling stake + capped-profit subsidiary"; from November 2024 onward, shifting toward a more typical for-profit structure
Public training rules	Claude Constitution fully public	System Card (partial) + Model Spec (from 2024)
Postmortem culture	Two Claude Code postmortems fully public	Various GPT regressions mostly handled internally
Data transparency	Transparency Hub + quarterly misuse [reports]	Usage Policies
Economic research	Full Economic Index report series + Programme	OpenAI Economic Research (from 2026, lagging Anthropic by 1+ years)
Safety-level triggers	RSP / ASL3 public triggers	Preparedness Framework (similar, but trigger cadence is opaque)
Public-interest funding	Gates Foundation $200M + Programme	(no equivalent scale)
CEO stance	Dario’s public essays + policy papers (2028 Two Scenarios)	Sam Altman’s X posts / personal remarks

Anthropic’s overall narrative:Institutionalizing mission / safety / transparency / governance in parallel as four things — After the November 2023 crisis, OpenAI gradually weakened its original governance design, while Anthropic went in the opposite direction (strengthening rather than weakening).

7. Related Wikis / Cross-Check Map

To see	wiki
Constitutional AI training methods + ASL levels + Welfare + Red Team	Complete Collection of Anthropic AI Safety / Alignment / Policy Research (Constitutional AI + RSP + Welfare + Policy)
Five major Economic Index reports + 81K survey data	Complete Anthropic Economic Index Collection (Measuring AI’s Economic Impact, 2025–2026 Series)
Enterprise products + strategic partnerships + four dimensions of M&A	Anthropic Enterprise / Industry Product Portfolio + Strategic Partnerships (2025–2026)
Claude model generations (Opus 4.7 / Sonnet 4.5/4.6 / Haiku 4.5 / Max Plan)	Evolution of Claude Product Releases (2024–2026 Complete Opus / Sonnet / Haiku 4.x Collection + Max Plan + Design)
AI industry judgments (model-company evolution / details of OpenAI’s November 2023 crisis / Anthropic’s $183 billion → $350 billion valuation)	AI Industry Judgments and Company Competition (Cross-Author Thematic Synthesis) / newtype observations on AI business, industry, and investment

8. Meta-information

Rewriting strategy: Reorganize the 9 sources according to "4 systems/institutions + public benefit + organization pages."Do not restate training principles / specific Constitution provisions(those change; look directly at the source), only explain structure + significance + industry comparisons.

Bilingual strategy: Section titles mix Chinese and English. Proper nouns such as Constitution / Transparency Hub / LTBT are retained, with Chinese explanations.

Industry criticism / skepticism explicitly listed (three LTBT questions, whether Constitutional AI is marketing, etc.), not smoothed over.