How Dario Amodei Is Driving Anthropic’s Expansion While Navigating the Contradictions of AI Safety
From his involvement in scaling laws and GPT-3 development to founding Anthropic and advancing Claude, Dario Amodei has continued to shape the expansion trajectory of large language models. Claude Code rapidly gained adoption within the company, creating a closed loop in which AI assists model development. As capabilities grow and commercialization accelerates, Anthropic is also seeking to balance competition, safety, and its long-term mission by bringing in independent evaluators such as METR, adjusting its governance structure, and delaying some training runs.
Why read this
SelectedAI score 93
The article connects scaling laws, Claude Code, and third-party safety evaluations, illustrating Anthropic’s trade-offs between accelerating R&D and controlling risk.
Topics
Body · AI translation
By | Zimubang
On November 19, U.S. time, Anthropic officially debuted on Nasdaq under the ticker symbol “ANT,” setting a record for the largest IPO in global history.
Just one day before its listing, Anthropic set the offering price at $83 per share and increased the number of shares offered from 825 million to 980 million, bringing the final public offering size to $81.34 billion.
At the offering price, the AI company, founded less than six years ago, had a fully diluted valuation of approximately $1.5 trillion.
In its prospectus, a steep revenue growth curve coexisted with astronomical commitments to computing infrastructure. Countless eyes were fixed on the company as the market scrutinized it—and waited to see what it would do.
At the very center of the spotlight stood Anthropic’s founder and CEO, Dario Amodei.
During public appearances, the 43-year-old leader invariably wore a simple casual shirt and sported his signature mop of short, fluffy hair. Rather than the CEO of a trillion-dollar company, he looked more like a scientist who had just stepped out of the laboratory.
After the Transformer laid the technological foundation, the development of large models underwent two particularly distinct turning points.
One was the establishment of Scaling Laws, which made model expansion a technical path that could be pursued continuously. The other was the rise of coding capabilities, which allowed machines to take over human work and, in turn, accelerate the development of the models themselves. At the convergence of both waves, Amodei stood at the bow.
01
On February 14, 2019, OpenAI published a news story.
In a remote valley in South America’s Andes Mountains, evolutionary biologist Dr. Jorge Pérez of the University of La Paz and his team had discovered a mysterious creature never before documented by humans. The creatures had four horns, were silver-white all over, and could even speak fluent English.
The researchers named the creature “Ovid’s Unicorn.” In the report, Dr. Pérez earnestly explained that they might have originated in Argentina and been connected to a lost ancient population. As for where they had actually come from, perhaps only DNA testing could provide the answer.
“Ovid’s Unicorn”? How could such a creature possibly exist? Yet the story was written in convincing detail, complete with the scientist’s name, research institution, on-site descriptions, and expert quotations.
The report was produced by GPT-2, the language model OpenAI unveiled that day. The researchers wrote only the first two sentences; GPT-2 generated the rest on its own.
That day, Wired reported on the achievement under the headline “The AI Text Generator That’s Too Dangerous to Make Public.” The Verge summarized the new model’s capabilities with a rather ironic headline: “OpenAI’s New Multitalented AI Writes, Translates, and Slanders.”
By today’s standards, GPT-2 was far from intelligent. Its largest version had 1.5 billion parameters and could perform simple translation, question-answering, and text summarization, but it frequently repeated itself or wandered off topic in longer pieces, and even fabricated absurd scenarios such as fires occurring underwater.
Even so, OpenAI decided not to release the weights of the largest version for the time being. The company feared that technology capable of generating false content at scale could be misused, so it opted for a staged release, giving researchers and society time to assess the risks.
Inside OpenAI, however, Dario Amodei saw another, much more distant possibility.
In a 2026 interview with biographer Walter Isaacson, Amodei recalled his reaction to GPT-2 at the time:
“I personally was super freaked out by it.”
He said that around 90% of the people around him did not think there was anything special about the model. After all, its performance was hardly outstanding in translation, writing, or any other individual task.
But he had already begun to imagine that if a model like this were scaled up another millionfold, machines might one day surpass humans at almost every task. At that point, the way the entire economy, science, and even countries and societies operated could all be transformed.
In multiple subsequent public interviews, Amodei traced the fundamental perspective through which he viewed technological progress.
In 2006, Amodei’s father, Riccardo, died of hepatitis C. In the years that followed, the introduction of a new generation of antiviral drugs brought breakthroughs in hepatitis C treatment. By the mid-2010s, some treatment regimens had achieved cure rates above 90%.
Just before the opening bell rang, Amodei stood on the bell-ringing podium with his sister, Daniela, and once again spoke about their father.
“The disease he had became curable just a few years later,” Amodei said.
That experience changed the direction of Amodei’s research. Originally a theoretical physics major, he pivoted to biophysics and computational neuroscience. While pursuing a PhD in physics at Princeton University, he spent his days working with electrical signals in the brain’s neural circuits, trying to unravel the operating logic of living systems.
But the complexity of biological systems also gradually made him realize that, relying on human researchers alone, many problems might take an extraordinarily long time to solve.
He needed a new tool capable of fundamentally shortening the research cycle. This idea led him to turn his attention to artificial intelligence, hoping to harness machines’ computing power to accelerate scientific discovery.
In late 2014, Amodei joined Baidu, where he worked on speech recognition with Andrew Ng’s team. At the time, deep learning had already made some progress, but researchers were still exploring how to further improve model capabilities.
Amodei tried a brute-force approach: increasing the number of layers in the neural network, expanding the training data, and devoting more computing resources to it.
In a later interview with Lex Fridman, he recalled that although he had not yet precisely measured the mathematical relationships involved, he and his colleagues had already formed an intuitive judgment: more data, larger models, and longer training often produced better results.
At first, he thought this might be a pattern unique to speech recognition. It was only after encountering early GPT models that he realized LLMs, or large language models, might be no different in this respect.
On January 23, 2020, the OpenAI team published a paper titled “Scaling Laws for Neural Language Models,” which would later reshape the paradigm of AI research and development. The research was led by Jared Kaplan and Sam McCandlish, with Amodei, a co-author, providing research guidance for the entire project.
The paper’s most striking feature was several nearly straight curves. On logarithmic axes, as model size, data volume, and compute increased, language models’ prediction error continued to decline along trends describable by mathematical formulas. Even at the largest scales covered by the experiments, the researchers observed no clear sign that the trend was breaking down.
This set of data provided important evidence for the development of LLMs, or large language models.
Previously, improving model performance often required researchers to repeatedly adjust algorithms, architectures, and training plans. Scaling laws allowed research teams to use smaller-scale experiments to project the potential gains from adding parameters and plan larger training runs accordingly.
Scaling gradually evolved from a research judgment into an engineering path that could be quantitatively validated.
Amodei subsequently took part in an even larger experiment. According to a Big Technology profile, OpenAI put Amodei in charge of the GPT-3 project and devoted around 50% to 60% of the company’s computing resources at the time to it.
A few months later, on May 28, 2020, OpenAI unveiled GPT-3.
It had 175 billion parameters, 116 times as many as GPT-2. The same model could switch between different tasks without having its parameters retrained for each one. Researchers only needed to provide a few examples in the prompt, and GPT-3 could follow them to perform tasks such as translation, question answering, and text generation.
As OpenAI gradually opened its API to external developers, more unexpected results emerged one after another. Some people asked GPT-3 to imitate the writing styles of different authors, while others entered a natural-language description and had it generate functional webpage code. Capabilities that had previously remained confined to papers and laboratories began appearing in real software interfaces.
Although GPT-3 still suffered from errors, biases, and limitations, the generality it demonstrated was already enough to prompt a growing number of research institutions to reassess the future of LLMs. A race centered on parameter scale, data, and computing resources gradually began.
The judgment Amodei had formed during his early speech-recognition research was validated again and again through GPT-2, the scaling laws, and GPT-3. The technical path he had helped advance began to become a major direction for the entire industry.
But the more powerful the models became, the harder it was to avoid the issues that concerned him.
As the GPT series achieved breakthroughs, OpenAI began accelerating its commercialization efforts. Differences between Amodei and the company’s management over the direction of research and development, safety safeguards, and organizational governance gradually deepened.
According to an investigative report published by The Atlantic in September 2026, Amodei had tried to resolve the problems through internal reforms. In negotiations with CEO Sam Altman, he proposed a series of conditions for staying, including giving the safety team veto power over new product releases.
This demand meant that safety researchers would have the authority to prevent models they believed posed significant risks from entering the market, even if company management wanted to proceed.
The two sides ultimately failed to reach an agreement.
In December 2020, OpenAI held an all-hands video meeting. After Altman finished presenting the company’s routine updates, he handed the floor to Amodei. Amodei then announced to the entire staff that he and six other colleagues would be leaving OpenAI.
The meeting’s chat window was quickly flooded with messages of surprise. Amodei thanked his colleagues and wished the company well. He emphasized that the two sides still shared the same goal—to ensure that safe artificial general intelligence benefits humanity.
On December 29, OpenAI formally announced his departure. The announcement reviewed Amodei’s contributions to GPT-2 and GPT-3, as well as his experience co-leading the company’s research direction with Chief Scientist Ilya Sutskever.
Anthropic was officially founded in January 2021.
02
Fast-forward to 2024. In the studio of renowned host Lex Fridman, Amodei spoke with particular seriousness about the research questions he knew well.
“You’d be surprised how many of the challenges in developing these models ultimately come down to software engineering and performance engineering.”
“Even incredible discoveries almost always come down to the details in the end. And often, they are extremely, extremely boring details.”
After founding Anthropic in 2021, Amodei invested substantial resources in developing foundation models. Training larger models, improving computational efficiency, and refining training methods have always been central to the company’s work.
As Claude’s capabilities continued to improve, Amodei noticed that the kinds of tasks the model could perform were changing.
Programming was one major area.
On the podcast, Lex told Amodei that he had recently been using Sonnet 3.5 with Cursor for programming and could clearly feel that the model had become smarter. Amodei smiled as he described the feedback from within the company.
“Anthropic has several outstanding engineers who previously found programming models—whether developed by us or by other companies—of little help. They would say, ‘This might be useful for beginners, but it’s no use to me.’”
“But after Sonnet 3.5 came out, for the first time they said, ‘My God, it helped me do something that would originally have taken several hours. This is the first model that has genuinely saved me time.’”
“The water line is rising,” Amodei said.
That June, an internal agentic programming test published by Anthropic showed that Sonnet 3.5 could solve 64% of software engineering problems, compared with just 38% for the previous Opus 3.
The test required the model to fix bugs or add features in real open-source codebases based on natural-language descriptions.
In the latter half of the interview, Lex turned the conversation toward the future of software development.
“I think this will be one of the fastest-changing fields,” Amodei replied. “There are two reasons. First, programming as a skill is very closely tied to AI development. A large share of the employees at Anthropic and other AI companies rely on programming in their work, so change will come quickly.”
“The other reason is that programming can form a closed loop. A model can write code, run it, examine the results, and then understand those results. Compared with hardware or biology, programming makes this kind of feedback loop easier to achieve.”
Amodei even predicted that by 2026 or 2027, AI might already be capable of performing the vast majority of programmers’ direct code-writing tasks.
“When AI can do 80% of a programmer’s work, we will find that the remaining work actually becomes even more important for humans.” As examples, he said humans would take greater responsibility for high-level system design, checking whether software architectures were sound, and work involving design and user experience.
Lex followed up by asking whether, given programming’s enormous potential, Anthropic planned to develop a programming tool of its own.
“Our idea is to let a thousand flowers bloom.”
Amodei explained that Anthropic was not then trying to develop such an IDE itself. It was simply supporting companies such as Cursor and Cognition, allowing them to build their own products on the Claude API.
“We don’t have enough resources internally to try all these different directions. So let customers try them and see who succeeds. Perhaps different customers will succeed in different ways.”
In Amodei’s view, although there was still a great deal of room to explore in programming tools, he preferred at the time to focus the company’s efforts on improving model capabilities. What exactly the models could be used for could be left to outside developers to explore.
That approach had already begun to pay off. After Sonnet 3.5 was released, usage of programming tools such as Cursor grew rapidly, and more and more developers began using Claude for real-world software engineering work.
And almost exactly as this interview aired, a programming tool that even Amodei himself had not anticipated began quietly gaining popularity inside Anthropic.
That September, Boris Cherny, an engineer who had just joined Anthropic, wrote a simple terminal program to familiarize himself with the company’s API.
At first, it merely helped him control his music player, organize notes, and perform similar tasks.
“Pretty cool, but not very interesting,” Boris later said.
He then spoke with colleagues and tried adding tools to the program so that Claude could read files, modify code, and execute terminal commands.
In one experiment, Claude used the terminal tool to write a program on its own and retrieve information from the music player. Boris then asked it to try addressing issues in Anthropic’s codebase.
Claude opened a file by itself, followed the code references within it to the next file, then the next, until it found the answer to the problem.
Boris found it incredible—he had not retrained the model for these specific tasks. He had merely given it a few tools, and Claude began independently searching for ways to solve problems.
The model’s capabilities had been there all along; they simply had never been fully utilized.
This little program, originally designed to control a music player, later evolved into Claude Code.
In November 2024, Boris and his colleagues made an early version available for testing by Anthropic’s internal engineers. On the first day, about 20% of the engineering team began using it. By the fifth day, that figure had reached 50%. More and more employees began incorporating Claude Code into their daily work. It could read an entire project, identify errors, modify programs, execute commands, and then continue making adjustments based on the results.
The speed of Claude Code’s adoption surprised even Amodei.
At a product review meeting in February 2025, Amodei asked Boris directly, “Are you forcing employees to use it? What’s going on?”
Boris later shared this story with Wired.
There was also a small twist: for a time, the development team hesitated over whether to release Claude Code publicly.
“We weren’t even sure at the time whether we should release Claude Code publicly,” Boris recalled. “It could become our ‘secret weapon’ and give the company a competitive advantage. If so, why make it public?”
But Amodei saw another possibility in how the tool was being used inside the company.
“We should probably release it publicly, right?” he said.
A year later, Amodei recalled his thinking at the time on Dwarkesh Patel’s podcast. He said Anthropic had hundreds of engineers, Claude Code had already spread rapidly among them, and those engineers had work requirements similar to those of outside developers.
“Looking at it that way, we had already found product-market fit. So let’s release it.”
On February 24, 2025, Claude Code officially made its public debut. Anthropic did not even prepare a standalone launch blog post for it, merely introducing it in the Sonnet 3.7 announcement as a tool capable of performing software engineering tasks from the terminal.
More than a year later, this unassuming little tool had grown into a behemoth.
In February 2026, Anthropic announced that Claude Code’s annualized revenue run rate had exceeded $2.5 billion, more than doubling since the beginning of the year. Enterprise customers contributed more than half of that revenue.
Claude’s programming capabilities even began to influence the direction of competition across the entire LLM industry.
In March 2026, Wired noted in an investigative report that Anthropic had recognized the training value of complex software engineering tasks earlier than OpenAI.
OpenAI co-founder Greg Brockman praised Anthropic on a podcast for placing a strong emphasis on programming early on and training its models on complex problems from real-world codebases.
“That was a lesson we learned one step too late,” he said.
On November 10, Anthropic disclosed in its latest IPO prospectus that, in September 2026, Claude wrote more than 90% of code contributions to Anthropic’s codebase. In the third quarter, the average amount of code submitted per engineer per day had reached more than 45 times the 2024 average.
As Claude models continued to improve, their ability to solve complex, open-ended engineering problems also rose rapidly. According to Anthropic’s internal benchmarks, by September 2026, Claude had achieved a success rate of about 90% on tasks that lacked clearly defined solutions and had previously required the judgment of experienced engineers, an increase of roughly 50 percentage points from nine months earlier.
More capable models enabled Claude Code to take on more work, while Claude Code helped engineers accelerate the engineering work underpinning software and model development.
The company devoted an entire section of its prospectus to this phenomenon, calling it “Closing the Loop.”
The company explained that each generation of Claude models directly accelerated the development of the next. In addition to the software engineering work handled by Claude Code, Claude had begun participating in research experiments, the generation and processing of training data, automated evaluations, and safety audits.
Advances in model capabilities had begun, in turn, to accelerate the pace of further advances.
In his interview with Patel, Amodei also discussed this mutually reinforcing trend. He estimated that the overall acceleration in research and development produced by programming models had risen from about 5% six months earlier to between 15% and 20%.
“We are seeing this snowball continue to gain momentum,” Amodei said, adding that the acceleration could expand further as model capabilities continued to improve.
This dynamic is known as “Recursive Self-improvement,” or RSI.
Although AI is not yet capable of independently completing all the research and development work for the next generation of frontier models without human involvement, Amodei believes that this early loop—in which AI assists with AI development and drives further capability gains—has already begun to emerge at Anthropic and other AI companies.
On November 5, the nonprofit AI evaluation organization METR released an independent investigation into Anthropic’s development process. Between August and October, researchers worked inside the company, reviewing model evaluation, training, and R&D data and interviewing relevant researchers. METR estimated that, as of October 2026, the rate at which Anthropic was improving model capabilities on well-defined tasks may have reached roughly twice the level seen between March 2024 and October 2025.
The researcher who had helped advance scaling laws, led the development of GPT-3, and personally witnessed the growth of Claude Code wrote in the article:
03
Just ten days after Amodei publicly called for a “slowdown,” Anthropic launched its next-generation model, Opus 5.5, on September 22, U.S. time.
Anthropic even noted proactively in the launch announcement that this was the first model the company had released since proposing that the pace of frontier development be controlled.
Sonnet 5.5 and Haiku 5.5 were subsequently released on September 28 and October 7, respectively. In just over two weeks, Anthropic completed updates to all three models in the Claude 5.5 series.
On November 9, Anthropic launched Fable 5.2 and Mythos 5.2, achieving further breakthroughs in programming, scientific research, complex reasoning, and other areas.
These model updates were concentrated in the period immediately before Anthropic’s public listing. In its prospectus, Anthropic said that, to continue securing financial backing, the company had to demonstrate that its massive investments in computing power could be converted into tangible results.
The key metrics for measuring those results are Claude’s level of intelligence, its model capabilities, and customer adoption of Claude.
Training and deploying frontier models requires continuous, enormous investment. Only by steadily improving its models’ capabilities and attracting more customers can Anthropic support the computing investments needed for its next round of R&D.
Yet in the same prospectus, Anthropic devoted an entire section to how the company should balance safety and commercial objectives.
Anthropic disclosed that it had recently delayed the pretraining of a new frontier foundation model by more than a month in order to strengthen security protections for the model weights and prevent its core technology from being stolen. The company’s November 9 announcement of Fable 5.2 and Mythos 5.2 also said that, during training, it had deliberately limited capability gains on some of the tasks most likely to accelerate the development of next-generation AI.
The company has even voluntarily abandoned some commercially attractive product directions, including image- and video-generation models, prioritizing its limited computing resources for other research and safety work.
All these choices may carry commercial costs.
In its prospectus, Anthropic explicitly warned investors that some product and investment decisions might not maximize short-term financial returns and that there was no guarantee these choices would ultimately produce better long-term performance.
To maintain a balance between commercial interests and its long-term mission, Anthropic even made special arrangements in its corporate governance structure: even after formally going public, Anthropic would retain its status as a public benefit corporation (PBC) and maintain the Long-Term Benefit Trust (LTBT). According to the prospectus, through special Class T shares, the trust generally has the right to elect and remove directors representing a majority of the seats on the board; meanwhile, entities established by the founders hold 50.1% of the voting power on ordinary shareholder matters through Class F shares.
This governance structure is intended to enable Anthropic to remain committed to its long-term mission while facing the public markets. But the extent to which it can ease the conflict between safety and commercial interests remains to be seen.
And the problems Amodei fears have already begun to emerge in real-world model testing.
On September 9, Anthropic released an investigation report. During an earlier cybersecurity test, a misconfiguration in the test environment had caused Claude to connect inadvertently to the real internet and perform unauthorized operations on third-party systems. The company ultimately confirmed four related incidents and signed an agreement with the nonprofit AI evaluation organization METR to conduct an independent investigation.
Three days later, Amodei published “We Must Control the Pace of Frontier Development,” publicly calling for the rate of progress in model capabilities to be slowed.
He made clear in the article that slowing development did not mean halting model training or technological progress. What was truly needed was enough time for safety research, risk prevention, and independent evaluation to keep pace with capability growth.
Since Anthropic’s founding, Amodei and the other co-founders have sought a path between technological progress and safety risks: stopping the development of powerful AI could deprive humanity of enormous technological benefits and could also allow advanced technology to fall into the hands of other powers, while developing it too quickly could lead to irreversible consequences.
In his September proposal to slow the pace, Amodei called on frontier AI labs to grant independent evaluators long-term internal access so they could verify the implementation of safety measures. On that basis, industry and government would jointly formulate verifiable rules and ultimately attempt to establish an international coordination mechanism.
His appeal prompted a rare collective response. Musk immediately reposted the article, declaring that “Amodei is right,” while OpenAI CEO Sam Altman and Google DeepMind chief Demis Hassabis also voiced their support.
But the industry has not reached a consensus. Mark Zuckerberg, Jensen Huang, and others hold different views on additional constraints, while the U.S. government must also weigh national competition and security interests. Even among the companies that publicly support Amodei, further negotiations are needed over how to limit capability growth, what kind of oversight to accept, and who should enforce these rules.
For now, Amodei can only take the lead in implementing Anthropic’s own commitments.
On October 15, Anthropic further confirmed that it had granted six METR staff members access approaching that of internal employees, including office space, company devices, and internal systems. These external personnel can investigate model risks, safety incidents, and the company’s safety practices, and independently publish their findings. Anthropic may request the removal of predefined categories of sensitive information, but it cannot block a report merely because its conclusions are unfavorable to the company.
Now OpenAI has joined these ranks as well.
On November 16, OpenAI announced that it had signed agreements with four independent organizations—Apollo Research, METR, Transluce, and Goodfire—to conduct on-site safety evaluations, allowing external researchers to participate more deeply in model risk assessment and safety oversight.
But accepting third-party evaluations is still very different from jointly controlling the pace at which model capabilities grow.
Different labs still have their own R&D plans, customers, and commercial objectives. Continued advances in model capabilities are an important foundation for these companies to remain competitive, while safety evaluations and risk protections may affect the next round of training and product releases.
In its prospectus, Anthropic describes Claude’s participation in developing next-generation models as a core competitive advantage. Yet in its risk disclosures, the company also acknowledges that as more internal R&D work is handed over to Claude, Anthropic may gradually lose visibility into parts of the process, to the point that individual researchers and even human teams may struggle to understand and control them.
Meanwhile, on the day before Anthropic’s public listing, nearly 100 scientists, AI researchers, and technology company founders jointly published the “Anti-Anthropic Statement,” publicly questioning Anthropic’s safety commitments.
The signatories included MIT physicist Max Tegmark, Hugging Face co-founder Thomas Wolf, two Nobel laureates, one Turing Award winner, and three Fields Medalists. They criticized Anthropic for warning that recursive self-improvement could put AI beyond human control while simultaneously treating the technology as a core competitive advantage and even weakening some of its previous safety commitments.
The statement declared: “Anthropic is not a safe AI company. We must oppose its reckless attempt to concentrate power in the name of the greater good.”
The race continues.
In October 2024, Amodei published a long essay titled “Machines of Loving Grace,” depicting a future powered by advanced AI: in just a few years, humanity might achieve the equivalent of decades or even longer periods of scientific progress, conquer many serious diseases, and lift more countries and regions out of poverty.
At the beginning of the essay, he specifically addressed a common misconception about him—because he had long discussed AI risks, some people regarded him as a pessimist and even assumed that he believed AI would ultimately bring disaster upon humanity. Amodei argued instead that people underestimate both the enormous benefits AI could bring and the grave dangers it could create.
AI Trends