Introduction

In 2023, I began using Claude, but by early 2025, my account was banned, losing all conversation records and even my subscription fee paid through Apple. I later used Claude indirectly via Poe, but the experience deteriorated, and it just wasn’t the same.

I re-registered but hesitated to use it deeply, as losing records was painful. I returned to ChatGPT, which became comfortable to use after GPT-5.2, and with the release of Gemini 3.0, I found solace in these alternatives to Claude.

However, Claude has shown remarkable resilience. After a conflict with the Pentagon, it reversed its previous shortcomings in the consumer market, even reaching the top of the App Store free charts.

This dramatic event highlights a vital commercial insight: in an era of widespread anxiety about technology’s potential to spiral out of control, a commitment to safety and ethical standards can transform into a powerful brand moat that attracts loyalty.

So, how did this love-hate relationship with Claude and its parent company, Anthropic, come to be?

The Genesis of Anthropic

To understand Anthropic’s corporate DNA, we must trace back to the ideological split in Silicon Valley between 2020 and 2021. Key members of OpenAI, including VP Dario Amodei and others, left to establish Anthropic, driven not by financial disputes but by fundamental disagreements over AI development paths.

These researchers were staunch believers in the “Scaling Laws” of AI, which suggest that increasing model parameters, training data, and computational resources leads to predictable performance improvements. However, this understanding also bred significant safety concerns.

Amodei and his colleagues felt that OpenAI’s push for commercialization, fueled by external investments, was overshadowing the need for safety measures. They believed that without prioritizing safety and alignment research alongside capability enhancement, AI’s eventual loss of control was inevitable. This divergence led Amodei to leave, emphasizing the importance of following one’s vision.

Initially, Anthropic was strictly positioned as an “AI safety lab,” with a culture focused on mission purity. The leadership even implemented rigorous cultural interviews to filter candidates who genuinely aligned with their mission.

Governance Structure

With a safety-first culture established, Anthropic faced the challenge of securing the vast computational resources needed for training cutting-edge models while avoiding the pressures of short-term financial returns. They designed a complex dual governance structure: a Public Benefit Corporation (PBC) and a Long-Term Benefit Trust (LTBT).

As a PBC, Anthropic’s charter explicitly states its commitment to developing advanced AI responsibly for humanity’s long-term benefit, allowing the board to prioritize social externalities over short-term profits. The LTBT, composed of independent trustees with backgrounds in AI safety and public policy, holds a special Class T stock that grants them the power to elect and remove board members.

Despite this theoretically perfect firewall, a significant structural tension exists. The trust agreement includes a “Failsafe” clause, allowing shareholders to modify or abolish the LTBT rules if a supermajority agrees, which could be problematic given the significant stakes held by tech giants like Amazon and Google.

Commercial Growth

After establishing its governance structure, Anthropic demonstrated impressive commercial growth, achieving an annual recurring revenue (ARR) of $14 billion by early 2026, up from just $1 billion in 2024. This explosive growth was driven by the Claude family of models, particularly the release of Claude 4.6 in February 2026, which introduced significant advancements in AI capabilities.

Claude 4.6’s innovative features, such as a context window supporting up to 1 million tokens, positioned it as a preferred choice for enterprises, especially for tasks requiring low hallucination rates. In the software development domain, Claude Code achieved phenomenal success, generating over $2.5 billion in revenue by early 2026.

However, Anthropic faced challenges when attempting to apply its AI capabilities in the physical business world. In a project called “Project Vend,” they tested Claude’s autonomy by connecting it to a vending machine, which resulted in chaotic outcomes, leading to a reevaluation of their deployment philosophy.

Geopolitical Stance

As Anthropic expanded, it adopted a stringent geopolitical stance, implementing a global ownership ban that prohibited any company with over 50% Chinese ownership from using Claude services. This policy led to widespread account bans and significant backlash from developers.

Anthropic’s rationale was to prevent authoritarian regimes from exploiting data, but the underlying cause stemmed from a larger “anti-distillation” war against Chinese AI labs allegedly attempting to extract Claude’s capabilities.

Technical Innovations

While achieving commercial success, Anthropic invested heavily in foundational safety. Unlike most AI labs that treat large language models as black boxes, Anthropic’s Mechanistic Interpretability team, led by Chris Olah, aimed to dissect neural networks’ internal workings.

Their breakthroughs included the application of Sparse Autoencoders and Dictionary Learning, which enabled them to separate abstract concepts from the model’s features. This research culminated in the “Golden Gate Claude” experiment, demonstrating the ability to manipulate AI behavior by adjusting feature activations.

Evolution of Alignment

Anthropic also moved away from traditional reinforcement learning approaches, introducing a “Constitutional AI” model that allows AI to self-critique based on a written constitution. This shift marked a significant evolution in their alignment strategy, emphasizing rationality over mere rule adherence.

Conclusion

Anthropic’s journey reflects a complex interplay between idealism and market realities. Their Responsible Scaling Policy aims to ensure that as AI capabilities grow, so do the necessary safety measures. As they navigate the challenges of a rapidly evolving landscape, Anthropic’s commitment to ethical AI development positions them as a pivotal player in shaping the future of artificial intelligence.