This is a Bloomberg (Bloomberg Originals) deep-dive profile of Anthropic's founding siblings — the 'visionary' older brother Dario Amodei and the 'operator' younger sister Daniela Amodei. It opens on the theme: Dario is an 'unlikely AI celebrity,' famous for warning the world about AI's risks, yet his company has become a frontier front-runner worth nearly a trillion dollars. The two have been close since childhood and are temperamentally complementary (Dario took calculus in middle school and math classes at Berkeley in high school, obsessed with sci-fi and understanding the universe; Daniela leaned toward reading and the humanities, and was an early Stripe employee). They joined OpenAI in 2016, where Dario proposed 'scaling laws' — the then-counterintuitive judgment that models get stronger simply by piling on data and compute, which gave rise to ChatGPT. He left in 2021 over disagreements with Altman about trust and values, and all seven co-founders remain at the company to this day (rare in the industry).

Claude's product philosophy is 'professional warmth': approachable but keeping its distance, its character trained via a 'constitution' and common human texts like the Universal Declaration of Human Rights, and they even consult religious leaders on how to instill cross-faith core values; the goal is to prevent the model from lying (hallucinating), deliberately deceiving, or producing harmful information. In the early Claude 2 era it was once 'over-nannying' (you just asked about the weather and it expressed deep concern), which researchers 'turned the knobs' to fine-tune into an appropriate range.

Commercially, Anthropic bets on coding and enterprise rather than 'splashy' consumer apps — Dario says bluntly that if the business model conflicts with values, you either betray the values or become irrelevant; social media / AI-video 'slop' chases addiction and ad-time, whereas enterprise use cases (biotech, pharma, academic research, energy) are naturally aligned with the mission of 'using AI to cure disease and cut costs while boosting efficiency.' The growth figures are stunning: 80x annualized in Q1, API volume up nearly 17x YoY, 8 frontier models delivered in a year, and a first profit. A key contributor is engineer Boris Cherny (creator of Claude Code / Claude Cowork, who before being recruited was making miso in rural Japan living a slow life) — he elevated AI coding from 'autocompleting a line' to 'a coding agent that can do everything,' and now a single engineer simultaneously directs anywhere from a few to thousands of Claudes writing code, with '100% of code written by Claude internally for at least half a year.' After Claude Cowork's launch the software sector evaporated about $285 billion overnight (the SaaSpocalypse). Dario thinks the pie will grow, but incumbents who can't see their moat or adapt will shrink or even fail.

Employment is the film's sore point. Dario reiterates that 'the next 1-5 years may cut half of entry-level white-collar jobs,' possibly producing the anomalous combination of 'high GDP growth + high unemployment/wage compression + high inequality': automation first makes people 10x more effective, but as it approaches 100% substitution, new things must be found for people to do. He forcefully rebuts charges from the likes of Jensen Huang that he 'conflates tasks with jobs' and that 'this is doom marketing that benefits Anthropic,' stressing that in long essays like 'The Adolescence of Technology' he spent five pages carefully distinguishing tasks from jobs and laid out responses from taxation to macro policy, only to be clipped into three-second videos and distorted. His answer to 'where can humans still go': physical/manufacturing work, human-centered relational work, and guiding roles that inject human intent into AI; he cites medicine as an example — AI will soon be great at diagnosis and ordering tests, but can't perform physical exams or 'bedside care,' so medicine will shift toward its interpersonal parts.

The second half turns to the sharpest issues of defense and power. Dario has long been anti-war (an anti-war type since his Caltech days), but 'the world has changed' — Russia-Ukraine and the Taiwan Strait convince him self-defense is necessary, so Anthropic became one of the first AI companies to sign with the Department of Defense and run on classified networks; it partnered with Palantir starting in 2024 and, together with OpenAI/xAI/Google, won a $200 million Pentagon contract in 2025, with Claude reportedly used in the operation to capture Venezuelan president Maduro. The conflict is that the DoD demanded 'guardrail-free' blanket use, and Anthropic drew red lines — refusing use for mass surveillance and autonomous weapons — and as a result was given an ultimatum and blacklisted by the Pentagon, with Trump and defense secretary Hegseth publicly calling Dario an 'ideological lunatic' and 'left-wing lunatic.' The heaviest passage in the film presses him: US officials claimed that with LLMs, the US military's daily strike-target count rose from 1,000 to 5,000; and in the Iran war, Claude reportedly did AI-assisted targeting via Palantir Maven, and one missile reportedly struck an Iranian girls' school, killing 150+ (mostly children). Dario responds that he 'doesn't know whether Claude was involved,' stresses that the principle they hold to is 'humans make the final decision,' and says it is precisely to prevent the scenario of 'AI deciding directly with humans unable to see it at all' that they are willing to 'gamble the company's future' to set limits.

Mythos is another undercurrent: a new model so strong it 'scares everyone,' able to find thousands of cybersecurity vulnerabilities and expose flaws in mainstream operating systems, which if fully released could be used to hack banks, pry open state secrets, and cripple critical infrastructure. An early tester said 'this is a super-weapon, it should require a gun license to use, please don't release it.' Anthropic gave it, via Project Glasswing, only to a handful of institutions (even the NSA, which had once blacklisted it, scrambled to use it), deliberately not fully releasing it and admitting 'huge commercial losses,' setting a benchmark that 'precisely as the leader you have the standing to turn the knob toward caution.' On 'why doesn't the government just take over such powerful private technology,' Dario admits this is a serious question that worries him too: historically nuclear weapons, the internet, GPS, and mobile phones mostly originated with government, and AI is the first powerful technology led by the private sector with government arriving late; he says he 'both fears the company holding it and fears the government holding it,' advocating a 'moderate, rational' middle path — one that needs mandatory pre-release testing and audits (the White House initially vetoed this, and on his first day Trump joined David Sacks to tear up Biden's AI executive order, but Mythos's national-security implications now have the White House wanting to 'guard the gate' again). He also holds firm on chip export controls to China (likening selling chips to China to 'selling nuclear weapons to North Korea,' openly at odds with some of his chip-maker customers, saying 'adults can cooperate and disagree at the same time').

The ending returns to the human and the risks: the attack on Sam Altman's residence left the siblings shaken, worried about the social hostility that AI's rising temperature brings; they treat social media's lessons on child welfare, mental health, and election integrity as a mirror, stressing 'we're lucky to be the second — we should proactively think through what could go wrong.' Dario self-assesses a 10-25% probability of civilizational collapse ('25% is too high, we need to push it very low,' analogizing 'you wouldn't board a plane with a 25% chance of crashing'), and proposes UBI and progressive taxation of AI companies to cushion the employment shock; his favorite book is 'The Making of the Atomic Bomb,' but he compares himself to Szilárd, who first conceived the chain reaction, and views Oppenheimer as a 'failure case that shouldn't have happened,' believing one can't rely on 'larger-than-life individual heroes' but on checks and balances everywhere. Daniela runs day-to-day operations (all executives report to her; she reports to no one), while Dario, every two weeks, gives a one-hour 'completely uncensored' unscripted talk to 3,000 employees to build consensus and shoulder the pressure.

(For AI investors: ① coding/enterprise is Anthropic's engine of profit and growth, and Claude Code-class agents are materially replacing software work and reshaping SaaS valuations; ② the 'capability threshold + safety threshold' of closed frontier models is becoming increasingly politicized, with defense, cybersecurity, and export controls all being hard variables in the supply chain; ③ the company proactively holding back Mythos and colliding with the Pentagon shows that the release pace of frontier models is increasingly driven by governance rather than pure commerce.)