Written by: Techub News Compilation

Introduction

In a deep interview program by Bloomberg called “The Circuit,” Anthropic co-founder and CEO Dario Amodei was interviewed for 70 minutes. At this time, with the Claude series models, especially Claude for Code and Claude Cowork that have ignited the enterprise market, Anthropic has transformed from a past “safety-first” challenger to a company valued near one trillion dollars and widely regarded as an industry leader. This conversation took place at a moment when the company was in the eye of a dual storm of public opinion and business, and Amodei rarely opened up, discussing grand topics ranging from personal pressure management and corporate governance to industry competition, geopolitics, and the future of humanity, providing a precious window for the outside world to understand the internal thinking of one of the most influential AI companies.

Summary

• Amodei compared the rapid development of current AI to a “relativity spaceship”: the external world is changing rapidly, while internally, decisions must remain as rational as a surgical operation to avoid oscillating between “not worrying at all” and “extreme panic.”
• He explained for the first time that the core reason for leaving OpenAI was not merely a technical safety disagreement, but a deeper issue of “trust and values,” emphasizing that Anthropic is practicing “responsible deployment” in its own way.
• Anthropic firmly chose the enterprise market rather than the consumer-level track because it believes that enterprise customers’ pursuit of long-term trust, safety, and positive values is more aligned with the company’s core values and can reduce fundamental conflicts.
• Regarding the powerful Mythos model not being made public, Amodei revealed it was due to strong demands from early partner companies, as the model’s capabilities in discovering and exploiting vulnerabilities had seen a “quantum leap” improvement, necessitating that protective measures mature before being gradually opened up.
• He reiterated his serious warning about AI potentially causing mass “white-collar unemployment,” but emphasized that the intention was to promote early planning solutions for the industry and society (such as skill transformation and macroeconomic policies), not to spread panic.

The “Relativity Spaceship” Captain in the Eye of the Storm

When asked how to cope with the suffocating speed of development in the current AI field, Dario Amodei provided a metaphor filled with physics imagery: it’s like riding a spaceship that accelerates away from Earth at relativistic speeds. “According to special relativity, you sleep for a while and wake up to find that two days have passed on Earth. Then you sleep again, and because the spaceship continues to accelerate, three days have passed on Earth... That’s how it feels.” He admitted that he has never been a person who sleeps much, but under exceptional pressure, he is learning “the art of relaxation and sleep.”

However, this exponential acceleration has not led to irrational panic. Amodei emphasized that while facing many clear and urgent challenges, he continues to address them while also thinking about how to prepare in advance. “Being paranoid or worried about what you’ll face upon waking is not constructive.” He drew from historical experiences of individuals handling high-pressure situations, pointing out that mature decision-making involves “not ignoring risks, not being complacent... but responding rationally.” He likened this ideal state to a surgeon conducting surgery or an officer commanding military action—maintaining basic calm while understanding risks. “Oscillating between ‘not worrying at all’ and ‘Oh my, we must panic today’ is a sign of immature decision-making.”

Amodei grew up in San Francisco, with a father who was a leather worker and a mother who worked in a library. He showed no interest in the internet revolution happening around him during his early years, being immersed in math, doodling, understanding the universe, and reading science fiction. He believes that the spirit of “not conforming, individualism, and being a little crazy” in San Francisco/Silicon Valley culture has profoundly influenced him. “It’s okay if experts disagree, if you have a coherent vision and worldview, you should pursue it.” Despite his many criticisms of Silicon Valley, he views this spirit of encouraging the exploration of “long-tail possibilities” as crucial.

The “Trust Gap” Leaving OpenAI and the Industry's “Carrot and Stick”

Regarding his departure from OpenAI and the founding of Anthropic, Amodei provided an explanation that transcended the common “safety disagreement” narrative. “Building powerful technology presents many difficult issues, and Anthropic faces many decisions daily where we are uncertain about what’s right or wrong. It is normal to have reasonable disagreements on safety issues,” he said, “but that alone is not enough to leave.”

The key lies in trust. “When you feel you cannot trust someone, when their values are not what they say, when they are dishonest, when they do not act for the reasons they claim, when you see disturbing patterns of dishonesty, it becomes difficult to continue collaborating with a company and trusting it.” Amodei summarized, “Ultimately, if you don’t share a common vision and do not trust each other, what’s the point in arguing? The solution is to go your separate ways.” He feels calm about the current situation: “We do things our way, and they do theirs. We will see who wins in the market and who wins in the court of public opinion. I think these are more persuasive than any dramatic story about who left for what reason.”

When discussing relationships with competitors, Amodei denied claims of mutual distrust within the industry. He cited DeepMind co-founder Demis Hassabis as an example, stating that they have known each other for 15 years and frequently exchange safety ideas. “My point is: first, some participants are more trustworthy than others. I think there are trustworthy participants beyond just Anthropic.” He proposed an industry collaboration perspective of “carrot and stick”: trustworthy participants need to unite to place untrustworthy participants in a position where they “have to adopt the same standards.”

“The positive side is motivating others, just like Demis and I motivate each other... that’s the ‘upward competition’ carrot side. Then there’s the stick side or the implicit stick, which is ‘these guys are doing the right thing, and those guys will look bad if they don’t do the right thing.’ We often see them grudgingly doing the right thing while trying to pretend they are doing something different.” Amodei believes this is precisely how to unite the industry and promote collaboration within it.

Value-Driven Business Choices: Why All in on the Enterprise Market

In the early days, while other companies focused on interesting, flashy consumer-level applications, Anthropic bet on programming and the enterprise market, achieving great success with Claude for Code and Claude Cowork. Amodei explained this was both a business decision and a values decision.

When founding Anthropic, the core has always been “doing the right thing.” But to realize this goal, funding is needed to support expensive model research and development; hence, the company must have a business model. The key is: will the business model conflict with the values? Amodei learned from the experiences of other companies: if the chosen business model fundamentally conflicts with the values, you will find yourself in trouble—either betraying the values or becoming irrelevant. “It’s far better to choose a business model that is compatible with the values.”

Looking at the mechanisms in the consumer world (especially social media) that encourage engagement and even addiction, along with the chaos in the AI video model field, Amodei saw the attention-maximization incentives driven by advertising revenue. The enterprise market, however, is different. “If we think about all the positive things that can be done with AI—curing previously incurable diseases, making energy cheaper and more efficient, aiding education, addressing health issues in developing countries, fostering economic growth—these all fundamentally belong to the enterprise domain.” Enterprise customers place a high value on long-term relationships and trust, which aligns very well with Anthropic’s goal of “deploying models in a positive and safe manner.” “This business model largely aligns with our values. While there are conflicts and difficult choices, the number of such choices is far fewer than in other cases.”

Regarding concerns that enterprise customers might easily switch models, Amodei acknowledged there is some inertia but never relied on it. “I think you want a better model, a better product.” He revealed that at least at the time of this interview recording, the growth curve had not shown a turning point; instead, it was on the rise.

“SaaSpocalypse” and Employment Shock: A Warning Rather Than a Doomsday Prophecy

After the release of Claude Cowork, traditional software companies saw their market values evaporate by $285 billion overnight, with traders dubbing it the “SaaSpocalypse.” Amodei believes it is very difficult to predict the specific impacts, but the competitive advantages of traditional software companies will diverge: the ability to write software quickly will disappear, but customer relationships, domain expertise, and unique knowledge will remain and even become more important. His advice is: never be complacent, list all the competitive advantages, and be aware that some will disappear while the relative importance of others will increase, while new competitive advantages may emerge. Those who adapt quickly will excel, while those who remain stagnant will struggle. His overall judgment is: the overall size of the software industry will grow rather than shrink, although there will be some major losers.

Amodei has attracted attention for warning that AI might eliminate half of entry-level white-collar jobs within the next 1 to 5 years, even being accused of “doomsday marketing.” He strongly refuted this, emphasizing that his complete narrative always includes discussions of solutions (such as “token taxes,” adjusting positions in collaboration with enterprises, macroeconomic policies, etc.), but social media tends to capture that “three-second doomsday clip.” He reiterated that his core message is: “This is something we should foresee, be concerned about, and actively respond to.” He admitted uncertainty about the specific numbers, but the degree of concern remains high, at the same level.

He envisioned a possible future scenario: on one hand, AI significantly improves productivity in positions like software engineers; on the other hand, there may also be cases where AI completes all work directly. The key lies in how society matches the new demand. He mentioned emerging roles such as “forward deployment engineer” or “AI solutions architect” within Anthropic, combining technical work with customer communication, which are in high demand. “Things will always adjust, but can the speed of adjustment keep pace with the scale of disruption? I don’t know.” He warned that the scale of disruption will be large, which is the significance of the warning—to prompt policy-making (whether at the internal company level or global macroeconomic level).

Looking five years ahead, Amodei pointed out several potential employment directions: jobs related to the physical world (such as building data centers and manufacturing), human-centered jobs (such as healthcare and customer service where human communication is still valued), and jobs that involve “guiding AI” (ensuring it aligns with human values and intentions). He firmly opposes simplifying the warning into a panic narrative of “revolution will rise,” “that is absolutely the outcome we want to prevent.”

Geopolitics, Military Applications, and the Mythos Model: Drawing Red Lines in Risk

Amodei openly supports chip export controls to China because he believes that China's lead in AI capabilities would be “highly detrimental” to the status of democracy in the United States and globally. Although some chip manufacturers disagree with his views, this has not stopped him from speaking out, even after signing more partnerships. “We are adults; we can cooperate on certain issues while having disagreements on others.”

Regarding AI military applications, Anthropic’s stance is complex and has sparked controversy. They were one of the first AI companies to sign contracts with the Department of Defense for operations in secure networks, but at the same time, they were excluded from some federal agencies due to insisting on certain safety measures. Amodei explained that, in light of the risks posed by Russia invading Ukraine and China's potential invasion of Taiwan, he is concerned about the aggression of a “revived authoritarian bloc,” believing that democratic countries need to defend themselves. This is the basis for their collaboration with two administrations. “We are certainly not in it for the money; that’s very troublesome... We do it because we care.”

However, caring also means that limitations must be set. He cited his statement in “The Adolescence of Technology”: “We should use this technology in every way, except for those that undermine our own values.” Their red lines are mass surveillance and fully autonomous weapons. “If democratic countries do those things, then winning for democratic countries isn’t worthwhile.” He explained this is why they were the first to collaborate with the Department of Defense while rejecting things some other companies are willing to do. “You should choose your principles and stick to them.”

Regarding their collaboration with Palantir, Amodei emphasized that its scope is carefully defined; they do not collaborate with ICE (Immigration and Customs Enforcement) or CBP (Customs and Border Protection) and do not consider having business in Gaza. He expressed deep sorrow over the tragedies that AI-assisted targeting might cause civilian casualties, but pointed out that this precisely highlights the importance of the principles they uphold—human beings must have the final decision-making power. “Imagine if we allowed fully autonomous weapons... that’s what we stand against.” He believes that, overall, the appropriate use of AI is more likely to prevent rather than provoke a world war, but if unchecked, it could increase the risk of misjudgments.

The handling of the Mythos model embodies Amodei's balance between capabilities and risks. This model has shown a “huge leap” in discovering vulnerabilities and turning them into exploitable attacks to the extent that early partner companies called it a “super weapon” and begged not to make it public. Amodei clarified that the goal is not to permanently lock down this model but to gradually open it up to a wider audience and ultimately release it to the public once strong cybersecurity measures are in place. They are concerned that existing protective measures (including those they released in Opus 4.7) might be circumvented, thus requiring time to strengthen defenses. He rebutted the claim that the capabilities of Mythos could be replicated using open-source models, using the discovery of 271 new vulnerabilities in Firefox as an example to prove its uniqueness. He emphasized that not releasing Mythos has caused significant damage to the company’s commercial interests, which is by no means a marketing gimmick. “We are doing our utmost to open up Mythos to new cyber defenders... this is a real challenge that we need to face as a society, rather than blaming it as cheap marketing.”

Governance, Future, and Trust: Keeping Balance on the Exponential Curve

Amodei argues that AI is the first powerful technology primarily built by the private sector rather than the government in history, a dangerous and unstable situation. He advocates for the need to establish checks and balances over the power of AI companies, such as the “Long-Term Benefit Trust” established by Anthropic, which has the authority to appoint and dismiss a majority of board members, essentially allowing it to fire him. This introduces a degree of “public governance” elements. At the same time, the government also needs to exert checks and balances through legislation (such as setting red lines) and the judicial system. “I fear both companies possessing this technology and governments having it. Companies need to provide checks on the government, and the government also needs to provide checks on companies.”

Regarding the “singularity” moment of AI self-improvement, Amodei believes this is not a single point in time but a continuous process. AI is already capable of suggesting architectures for the next generation of AI, and the productivity improvement rate is growing exponentially. The key lies in assessing at each point whether it is necessary to slow down or strengthen control. “The correct and wise response to this technology is to say: we will not panic. Our countermeasures will smoothly upgrade with technological capability.” He criticized the extreme oscillation between “opposing all AI regulation” and “advocating for nationalization” as a sign of being “not serious.”

Amodei once estimated the risk of AI causing civilization collapse to be 10%-25%; he emphasized that Anthropic's actions aim to reduce rather than increase this probability. He likened the company to an airline striving to be ten times safer than other airlines but cannot guarantee it will never crash. “25% is too high; we are striving to greatly reduce this probability, and that is the goal.”

Finally, regarding why the public should trust him and his company, Amodei acknowledged that starting from distrust is rational, given the behavior of Silicon Valley in recent years. He believes trust must be earned through actual actions. He cited Anthropic’s record of “walking the talk”: not releasing Mythos has harmed commercial interests; proactively cutting off model access to mainland China cost hundreds of millions (at the time a substantial portion of revenue); delaying the release of Claude, etc. “We are not perfect; we will make mistakes... but fundamentally, we have an honest and sincere vision of how to do the right thing, and we are striving to execute that vision.” People need to take a long view of history to judge which assumptions align best with these actions.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。