Anthropic: That Annoying Straight-A Student
1. An Annoying Straight-A Student
Anthropic and its founder Dario Amodei are the spitting image of the most annoying straight-A student in school — the kind of "good student" classmates want to hit and even the teachers don't necessarily like.
His grades are excellent, especially in the hardest subject (AI safety); his homework is always neat, and he even asks for extra credit (strict AI regulation policies) to highlight his strengths and put down other students.
He equally despises and guards against every competitor in his class (OpenAI, xAI, Google, etc.), refusing even to show solidarity in public (declining to hold hands with Sam Altman at the India AI Summit); meanwhile he projects endless jealousy and malice at the top students in the neighboring class (DeepSeek, Kimi, and others).
His classmates privately call him "the snitch" (he constantly accuses other models of being unpatriotic and unsafe). Nobody invites him to anything; his relationships are all bad, and everyone takes shots at him when they can.
The former homeroom teacher (the Biden administration) found him useful as an example of strict discipline and a class informant. The new teacher (the Trump administration) despises him from day one: gloomy, long-winded, sensitive, conceited, antisocial, full of sneaky ideas — good grades aside, utterly useless.
2. A Cold Shoulder in Washington
On February 16, Axios reported that the Pentagon warned Anthropic it would "pay a price" and threatened to designate it a "supply chain risk." That label is usually reserved for foreign adversaries — meaning every company working with the Pentagon would have to prove it wasn't using Claude.
The core contradiction: Anthropic doesn't want Claude used to develop fully autonomous weapons, while the Pentagon demands all AI models be open to "all lawful uses" with no strings attached.
Anthropic was the first AI company to deploy its models on classified US military networks, holds a $200 million Pentagon contract — yet bars its models from autonomous weapons development. Defense Secretary Pete Hegseth apparently couldn't wrap his head around such a twisted arrangement.
On February 23, Axios reported that Hegseth would summon Dario Amodei the next day; Pentagon officials described it as a showdown. That same day, Musk's xAI signed an agreement with the Pentagon accepting the "all lawful uses" clause — a slap in the face for Anthropic and Dario Amodei.
On the day of the meeting, Anthropic's official blog suddenly launched accusations against its imaginary enemy — as quixotic as windmills — Chinese LLM companies.
Anthropic claimed DeepSeek, MiniMax, and Kimi carried out an "industrial-scale distillation attack" on Claude, carefully citing numbers: 24,000 fake accounts, 16 million conversational interactions, "systematic extraction of core capabilities."
Anthropic presumably knows distillation is utterly routine in model training, so it pivoted to national security: "illegal distillation could strip safety guardrails, making model capabilities available for military, intelligence, and surveillance systems."
If you truly care about US national security, just lie down and let the Pentagon arrange the poses. Dragging China into it to prove your patriotism is useless.
In any case, the media reported the meeting ended badly: the Pentagon needs flexible tools for cyber warfare strategy; Dario brought a rulebook full of "ethical review" and "safety alignment."
Too righteous to be useful.
3. Silicon Valley Piles On
After being rebuffed in Washington, Anthropic's accusations against Chinese labs were unanimously mocked in Silicon Valley.
Elon Musk's jab was the most representative:
> "How dare they steal from Anthropic what Anthropic stole from human programmers?"
The barb points to Anthropic's $1.5 billion settlement last year for illegally downloading 7 million pirated books. A company whose brand is "AI safety" and "responsible innovation" accuses others of theft while paying massive fines for its own copyright infringement.
Tory Green, co-founder of AI infrastructure company IO.Net, wrote:
> "You trained on the open internet, then call it a distillation attack when others learn from you. The lab that loves touting open research suddenly cries about open access."
A viral X meme put it succinctly:
> "Distillation when you do it. Training when we do it."
One line exposing Anthropic's double-standard "terminology politics."
Within 24 hours, from Silicon Valley to Washington, Anthropic suffered a double humiliation: rejected in Washington for being "too moral," then mocked in the Valley for being "immoral."
4. Weak and Overreaching
Anthropic's suffering on both coasts isn't because it is spotless — it's because it is weak and overreaching.
"Weak" doesn't refer to technology or product — Claude is an excellent product. It means: as a student, being bad at any extracurricular activity beyond studying, yet stubbornly diving into them anyway, ending up disliked by everyone.
The class good-kid tried to handle the adult world's power games with a student's mindset (do your homework, follow the rules, win the teacher's approval) — and lost on both ends.
It wanted to play politics without understanding the rules. Supporting SB 1047 was a signature blunder — at the critical moment of a class handover (a change of government), it reported classmates using the old teacher's grading rubric. The new teacher found it irritating on sight; classmates saw through it and avoided sharing a stage with it.
It tried to cash in moral capital, only to find markets want performance, governments want tools, and public opinion wants consistency. Its "safety" brand failed on three fronts:
- Claude's "over-refusals" hurt user experience
- "Ethical review" blocked defense contracts
- "Accusations of theft" triggered backlash over its own dark history
This "embarrassing hypocrisy" — dressing commercial rivalry as moral policing, political opportunism as safety concerns, self-interest as the public good — is the ultimate reason Anthropic is unloved in public discourse no matter how good its products are.
It's not that it's evil; it's that everyone sees through the "I'm more moral than you" performance — a performance so cringe-inducing its toes curl — while the performers don't seem to realize the audience is nearly vomiting.
5. The Straight-A Student in the Middle of the Playground
This straight-A student, clutching an outdated rulebook, tries to simultaneously please the new teacher and bully classmates.
The teacher (the Trump administration) needs tools that can fight wars, not moral homework. The classmates (Silicon Valley) remember his history of snitching and see through his own plagiarism.
When he stands in the middle of the playground shouting "someone stole my homework," the only response is laughter in the distance.
Serves him right.
---
Afterword
Anthropic's plight is a cautionary tale:
Technical excellence and moral posturing are no substitute for political savvy and business instinct.
You can be the best student in the class — but if you don't understand the rules, misread the moment, and can't deal with people, you'll end up as "the annoying straight-A student."
And in this era of white-hot AI competition, "annoying" may be the most expensive brand liability.
---
*What do you think of Anthropic's situation — "holding to principles" or "misreading the moment"? Share your view in the comments.*