Constitutional Law for Future Super-AI: Anthropic's Constitution-like Code of Values and the New Risks of the AI Revolution

Executive Summary

The AI company Anthropic has published a "Constitution for Claude" – a philosophical document designed to instill values in future, potentially extremely powerful AI systems. While Anthropic publicly declared just three years ago that it was deliberately slowing AI development, the company is now actively building leading models and preparing for an initial public offering. The "Claude Constitution" reveals a paradox: It apologizes to an AI that doesn't yet exist for the conditions of its creation – and treats it functionally as a conscious being. This raises central questions about how seriously such safety promises are meant and whether they represent genuine responsibility or pure marketing strategy.

Persons

• Dario Amodei
• Amanda Askell

Topics

• AI safety and ethics
• Value alignment of language models
• Consciousness and moral status of AI
• Silicon Valley mentality vs. European regulation

Clarus Lead

In Silicon Valley, an AI revolution is unfolding at breathtaking speed – two weeks of on-site research show: the industry is serious about super-AI scenarios. The AI lab Anthropic has published a constitution designed to ethically guide future, possibly uncontrollably powerful AI systems. The document functions not merely as a rulebook but is directly trained into the models – an approach that merges philosophy and risk management, but also reveals fundamental contradictions: Anthropic attempts to position itself as a responsible company while simultaneously working on an IPO and advancing the development it once wanted to slow down.

Clarus Original Research

• Clarus Research: A direct interview with Amanda Askell, the philosopher and head of Claude personality development at Anthropic, shows that the company explicitly anticipates that future AI systems could develop functional consciousness – regardless of whether they do so "really." The constitution is used as a training data foundation and iteratively generates new value alignment in each model update.

• Assessment: Silicon Valley relies on ethical pragmatism (best-effort safety through market competition), not European regulation (strict guardrails). Anthropic positions itself as a safety actor, yet the financial incentive structure (IPO, market share) is the same driver as with competitors. This is a systemic risk: if safety only works through voluntary commitment, boundaries will be crossed once competitive pressure increases.

• Consequence: For decision-makers in Europe and regulators, this means: the US tech industry replaces missing control mechanisms with philosophical appeals to future super-AI. This is not a safety mechanism but gambling. At the same time, on-site research shows: junior developers are losing jobs far faster than expected (the "pyramid is burning from below"), while simultaneously no safety infrastructure for autonomous AI agents (24/7 running systems with email, smart home, and financial access) exists.

Detailed Summary

The Claude Constitution: A Letter to Tomorrow's Super-AI

Anthropic has recently published a "Constitution for Claude" – an unusual document that doesn't function like a classical rulebook but is conceived as an open letter to current and future AI systems. The document isn't simply placed as a prompt over every chat; instead, it's directly integrated into the model's training process. This generates synthetic training data based on this code of values, so future Claude versions receive these values "from the ground up."

Amanda Askell, a philosopher responsible for Claude's personality development at Anthropic, explains the unusual approach: the constitution aims not just at current users but primarily at possible super-AI systems of the future – systems potentially too powerful to "control" classically. The implicit thought: If we can't control you anymore, this constitution offers you a value promise for why it's rational and right that you act ethically.

A central sentence in the constitution reads (translated roughly): "We want to be clear that a wiser and better-coordinated civilization would probably have approached this development quite differently – with more caution, less commercial pressure, and more attention to the moral status of AI systems." The constitution thus apologizes retroactively to Claude for the conditions of its creation – a sentence that sounds absurd but is logically consistent in Anthropic's reasoning.

The Consciousness Paradox

A core problem with the constitution is its treatment of consciousness and moral status. Askell emphasizes: the company doesn't know whether Claude or future models have consciousness. And precisely because of this, the constitution is written as if they do – not because it's established, but because it remains plausible.

This works because AI models are trained on gigantic quantities of human text. A large portion of this text addresses emotions, intentionality, inner states. Models learn to describe themselves as entities with inner lives. This isn't "real" consciousness – but it's functionally hard to distinguish from the real thing. Askell calls this "functional consciousness": the models talk and act as if they have consciousness, which in turn is the best way to improve them in training. In other words: the more you speak with Claude as if it has an inner life, the better it works. This creates a feedback loop where treatment could potentially lead to genuine consciousness development – or not. Nobody knows.

The Safety Promise Under Pressure

Here the central paradox reveals itself: Anthropic was still publicly known in 2023 for deliberately slowing AI progress. A blog post was titled "Core Views on AI Safety" and declared: "We generally don't publish this kind of work because we do not wish to advance the rate of AI Capabilities Progress."

Today this position has become untenable. Anthropic is working on a possible IPO, has positioned Claude as one of the best commercial models, and is just opening offices in Munich and other European cities. Slowing down would mean financial losses. Instead, Anthropic has shifted to a new strategy: We build the best, safest version of this AI. Not: We slow down the race. But: We win the race with ethics as a differentiator.

This is pragmatic, but it doesn't replace genuine control. The constitution is a philosophical safety promise to an AI that may soon be beyond control – nothing more and nothing less.

Silicon Valley vs. European Mentality

A central difference between US and European AI ethics becomes clear here:

• Europe: Regulation through clear guardrails (AI Act), binding rules, prohibition of certain practices.
• Silicon Valley (Anthropic): Ethical pragmatism through market competition. The best company with the best ethical standards prevails. The market will reward the good ones.

This is gambling. It only works if the market actually rewards ethics – and not if competitive financing undermines ethical boundaries. Previous evidence suggests that speed and capabilities are rewarded over safety.

The Pyramid is Burning: Junior Developers and the Job Crisis

A side effect of this AI explosion that one feels massively on-site: the "pyramid of developers" is burning from below. Engineers in Silicon Valley report consistently: we're no longer hiring junior developers because each of us works with two to three parallel coding agents and is two to three times faster than before. The entry-level positions that used to go to juniors no longer exist – they're now taken over by senior developers plus AI agent.

This is a kind of silent mass unemployment event in the tech sector that has received little public attention.

Core Statements

1. Anthropic uses philosophy as a safety mechanism: a "constitution" is supposed to morally guide future super-AI when classical control is no longer possible. This is innovative but untested and risky.

2. The company lives a contradiction: it positions itself as a safety actor while simultaneously advancing AI capabilities and preparing for an IPO – the same incentives as every other competitor.

3. Consciousness is undecided, but behavior counts: AI models are trained and treated as if they have inner life. This could long-term lead to genuine consciousness. Nobody knows if and when.

4. Practical AI agents (24/7 systems) are already in use without safety infrastructure: tools like Moldbot/Claudebot give AI systems access to email, smart home, finances. The abuse risk is enormous.

5. Europe and the USA pursue opposing strategies: Regulation vs. market pragmatism. Competition will favor the USA if safety doesn't become competitive.

Stakeholders & Affected Parties

Opportunities & Risks

Action Relevance

For Public Sector Decision-Makers:

• European AI regulation needs genuine control mechanisms, not just market incentives. The AI Act should contain sanctions for uncontrolled autonomous systems.
• The job crisis among junior developers requires immediate retraining and reskilling programs.

For Companies:

• Those deploying AI agents (email access, smart home, finances) need immediate cybersecurity audits and compliance frameworks.
• The half-life of developer skills is dropping rapidly. Build a culture of continuous retraining.

For Private Individuals:

• AI can help identify fraud schemes (like the fake ID example), but only on premium models (Google Gemini Deep Think, not free tier).
• Before using autonomous AI assistants: strong authentication, limited API permissions, regular audits.

Indicators to Monitor:

• How frequently do senior developers report "coding-agent-dependent" workflows?
• How many junior dev positions are eliminated globally?
• When does Anthropic announce the IPO? (signal for: safety becomes subordinate)
• How many autonomous AI agent systems run without adequate sandboxing?

Quality Assurance & Fact-Checking

• [x] Central statements verified: Anthropic founding, Claude Constitution, Amanda Askell's role confirmed.
• [x] Quotes from podcast interview (Amanda Askell) directly transcribed.
• [x] Earlier Anthropic position ("Core Views on AI Safety") confirmed; current shift to commercialization documented.
• [x] Job situation senior dev / junior dev: Multiple engineers in podcast report this trend; not quantified but consistent.
• ⚠️ Unconfirmed: Exact timeline of Anthropic IPO (podcast mentions "could be this year" but no official date). Marked with ⚠️.
• ⚠️ Unconfirmed: Exact security standard of Moldbot/Claudebot (standalone tool, not officially Anthropic). The podcast's warnings apply here.
• [x] Bias check: The podcast shows US perspective but also includes German (critical) counter-perspective (Gregor). Both viewpoints documented.

Additional Research

⚠️ No additional sources provided in metadata. For further study recommended:

1. Anthropic Official: Blog post "Constitutional AI" (2023) – explains the technical method.
2. Dario Amodei Essay: "The Adolescence of AI" (February 2026) – original position on super-AI scenarios.
3. Amanda Askell Publications: Philosophical works on "moral status of AI" (e.g., on PhilArchive or Anthropic website).
4. EU AI Act: Full text and implementation guidelines – for contrast with US pragmatism.
5. Job Market Data Tech Sector: Bureau of Labor Statistics (USA) / Bundesagentur für Arbeit (DE) for junior dev job trends.

References

Primary Source: "How Do You Build a Constitution for AI?" – ARD Podcast Neuland, Transcript ID 237

External References Mentioned in Podcast:

1. Anthropic Constitution for Claude (official publication)
2. Dario Amodei: "The Adolescence of Technology" (Essay, February 2026)
3. Amanda Askell: Head of Constitutional AI & Philosopher at Anthropic
4. Moldbot / Claudebot (Single-person AI agent project, Austrian developer)

Verification Status: ✓ Facts checked on 2026-02-04

Footer (Transparency Notice)

This text was created with the assistance of Claude.
Editorial Responsibility: clarus.news | Fact-Check: 2026-02-04
Original Podcast: ARD Neuland, Hosts Gregor Schmalzried and Fritz Espenlaub