Two red lines
Juan narrates the Anthropic defense-contract story as an ethics teardown. Anthropic set two non-negotiable conditions to work with the US Department of Defense: zero domestic mass surveillance (no reading citizens' chats) and no autonomous weapons without a human in the loop, a person always has to pull the trigger. The DoD refused, insisting on full access or else, and Anthropic was branded a national-security supply-chain risk and lost the contract. Then, hours later, OpenAI announced it had taken the deal 'on the same conditions', which makes no sense: why would the government accept from one company what it rejected from another? The suspicion Juan raises is that either those conditions weren't really honored, or a Patriot-Act-style law makes them hollow, since anyone labeled a terrorist stops being a protected citizen, a dictator's tool. He frames it through motive: Amodei left OpenAI precisely to steer AI's direction and pushes hardest for regulation, while for Altman, OpenAI is his baby, cash-burning and in debt, so he'd take an unethical deal to keep it alive, the same week a hundred-billion investment lands. Meanwhile the labor signal splits: Anthropic keeps hiring while Jack Dorsey fires four thousand by tweet and the stock jumps.
// trace: where this idea came from
- ↳ video diary @ 39:37 (primera línea roja, cero vigilancia masiva doméstica)
- ↳ video diary @ 41:09 (segunda línea roja, cero armas autónomas, el humano aprieta el gatillo)
- ↳ video diary @ 42:05 (horas después OpenAI dice que aceptó el trato con las mismas condiciones)
- ↳ video diary @ 43:01 (la sospecha, la ley patriota deja espiar a cualquiera declarado terrorista)
- ↳ Entry 236-1: Every move leaks (el hilo de vigilancia que esto lleva a escala de estado)
Juan tells the Anthropic defense-contract story as an ethics case. To work with the US Department of Defense, Anthropic laid down two non-negotiable red lines. First, zero domestic mass surveillance: with every citizen’s chats in hand you could trivially pull up anyone and read what they’re asking, from affairs to illnesses to how to build a bomb, and that data is too sensitive to touch ▸ 39:37. Second, no autonomous weapons without human control: the model must never be the one deciding who lives and who dies, a human always pulls the trigger ▸ 41:09. The DoD refused, demanding full access or consequences, and Anthropic was branded a national-security supply-chain risk and lost the contract ▸ 41:37.
¿por qué el gobierno aceptaría de uno lo que le rechazó al otro? →
Then, hours later, OpenAI announced it had taken the same deal on the same conditions, which is where the suspicion starts: why would the government grant one company what it just denied another ▸ 42:05? Either those conditions weren’t honored, or a Patriot-Act-style law makes them hollow, since anyone the president brands a terrorist stops being a protected citizen and can be surveilled anyway, which is how one man becomes a dictator ▸ 43:01. Juan reads it through motive. Amodei left OpenAI precisely to hold power over AI’s direction, and is among the loudest voices pushing law to catch up with the technology ▸ 45:42. For Altman, OpenAI is his baby, cash-burned and in debt, so he’d accept the unethical deal to keep it breathing, and days later a hundred-billion investment lands, because alignment with the government de-risks you ▸ 47:49. The labor signal splits the same way: Anthropic keeps hiring while Jack Dorsey fires four thousand people in a single tweet and the stock climbs…