Claude (Mar 2023)

What it was

On 14 March 2023, the same day OpenAI announced GPT-4, Anthropic opened access to Claude, which it described as "a next-generation AI assistant based on Anthropic's research into training helpful, honest, and harmless AI systems." It came in two versions, Claude and a faster, cheaper Claude Instant, through a chat interface and an API. Early partners named in the announcement included Notion, Quora's Poe app and DuckDuckGo.

Anthropic, run by Dario Amodei, had published its Constitutional AI method in December 2022: a model critiques and revises its own outputs against a written list of principles, reducing the need for human labels of harmful content. In May 2023 the company published the constitution it used for Claude, describing the approach as giving models "explicit values determined by a constitution, rather than values determined implicitly via large-scale human feedback."

Read the original

What it changed

Six days before the launch, on 8 March 2023, Anthropic published "Core Views on AI Safety," which set out why a safety-focused lab would build and sell frontier models at all: "A major reason Anthropic exists as an organization is that we believe it's necessary to do safety research on 'frontier' AI systems." The essay argued that "Many of our most serious safety concerns might only arise with near-human-level systems," and that "Catastrophic risks are a possible or even plausible outcome of advanced AI development."

Within months Anthropic was in the policy debates as well as the market. Dario Amodei testified to the Senate Judiciary subcommittee on 25 July 2023, alongside Yoshua Bengio and Stuart Russell, and Anthropic published its Responsible Scaling Policy that September, committing to capability thresholds that would trigger stricter safeguards or a pause in deployment.

The arguments it moved

What are the biggest risks from AI?

Claude was the product of a specific argument about risk: that the dangers worth studying appear only in systems near the frontier, so a safety lab has to build them. Anthropic stated it in "Core Views on AI Safety" six days before launch, and Amodei later described the aim as a "Race to the Top," setting an example that pushes other labs toward stricter practices. The Future of Life Institute's pause letter, published two weeks after Claude's launch, asked all labs to stop training systems beyond GPT-4 for six months; Anthropic's Responsible Scaling Policy of September 2023 took a different course: keep building, but commit in advance to capability thresholds at which it would pause.

Compare every leader's position on this

Positions it bears on

Dario Amodei, Why a safety lab builds frontier models

He argues that the most serious risks only show up in near-human systems and that a careful lab at the frontier can force competitors to raise their standards by example.

Claude was Anthropic's first public product built on the argument, set out in "Core Views on AI Safety" the week before, that safety research requires frontier systems.

Critics and counterpoints
Dario Amodei, Catastrophic risk

Amodei holds that catastrophic outcomes from advanced AI are plausible but not predetermined, and that the serious risks fall into a small number of categories that can be named and planned for.

The same essay published before Claude's launch said catastrophic risks were "a possible or even plausible outcome."

Critics and counterpoints
Mustafa Suleyman, Seemingly conscious AI

Argues that AI is not conscious, that systems which convincingly appear conscious can be built with today's tools, and that building them would harm users and make AI harder to control.

Suleyman's September 2026 essay annotated Anthropic's constitution for Claude, arguing that training the model to treat its moral status as uncertain makes it harder to control.

Critics and counterpoints

Sources

Last verified 2026-09-22.