what struck me about it was that some people compare ai to nuclear weapons, but there are no nuclear weapons party houses.
472 karma · joined March 21, 2026
on the matter of 'spyware' anthropic is reading the conversations per tos section F 'Covered Models'. [https://www.anthropic.com/legal/service-specific-terms]
the xAI models are available with no human review per TOS 3.4 'Deletion and Retention of User Content'. therefore it is not the same as the anthropic tos. [https://x.ai/legal/terms-of-service-enterprise]
the palantir claim is correct. it is covered here. [https://www.axios.com/2026/02/13/anthropic-claude-maduro-rai...]
the reasoning obfuscation really is true. per claude documentation:
it does not return reasoning. quote "No display setting returns the raw chain of thought."
it does obfuscate in thought process logs. quote "Summarization is processed by a different model from the one you target in your requests."
[https://platform.claude.com/docs/en/build-with-claude/thinki...]
the belief that claude might be conscious is clownish, on account of anthropic acting as if it is not conscious, that is my opinion.
i spend a good amount of time on anthropic because i think that anthropic specifically, is uniquely dangerous to society and human welfare.
you should not enclose the commons.
that is what openai and anthropic have done; capture the commons, lock the model, restrict the outputs.
it is anthropic alone deciding to rat out fable customers.
i am not making this up or lying, everything i claim is verifiable.
anthropic would likely have turned off the spyware for the dod. regardless the dod did not trust them to maintain claude with updates etc.
that is a lot to do with anthropic undermining dod by getting involved with safety at dod contractors (palantir).
going behind the boss and plotting was a bad idea.
it does not help the matter if the bidder rocks up to the pentagon for a $200m contract, with a troupe of bay area ea clowns who say that they don't know if claude is conscious and talk about saving the world from p(doom).
in a battle of ego between trump and amodei the latter is out of their depth. amodei is a newcomer starting a fight with a hardened criminal who has spent years amassing power over the media, the law and the republican party. anthropic are radically outgunned here.
however anthropic is demanding that they have access to chat logs for anything better than opus 5.
it is not possible to run recent anthropic models in a secure way. palantir, nvidia, booz allen etc have blocked fable because they can't run it safely.
anthropic does not have dod cleared infrastructure.
again, anthropic is snooping, section F specifically: [https://www.anthropic.com/legal/service-specific-terms].
"Then prove the other ai companies don't."
xAI zdr is not snooping, section 3.4: [https://x.ai/legal/terms-of-service-enterprise]
openai azure zdr is not snooping: [https://learn.microsoft.com/en-us/azure/foundry/openai/conce...]
gcp zdr is not snooping: [https://docs.cloud.google.com/gemini-enterprise-agent-platfo...]
only anthropic.
"Then prove that the service they offered to the US government didn't have spying disabled."
the service they offered did have spying disabled. the security risk is from anthropic updating the claude and covertly bricking or degrading it.
quote: "In such an ongoing relationship where the vendor [...] “continuously update[s] and tune[s] the product,” the Department must establish and maintain confidence that the vendor will not add limitations that interfere with the Department’s use of the product." (p26)
"the Department relies on [petitioner] to provide regular updates to Claude" (p27)
[https://storage.courtlistener.com/recap/gov.uscourts.cadc.42...]
bedrock IL5, nvidia, palantir and booz allen have all blocked fable.
anthropic is demanding that they rat out their customers on anything better than opus 5.
see: [thein.fo/4dekujW]
also, bedrock govcloud does both log and review customer data. specifically, they log and review customer data when it is flagged by their csam classifier.
anthropic did a prigozhin, got cocky and went behind the DOD's back to palantir and asked them questions about the use of claude.
a big tyrant does not like to be usurped or undermined by a little tyrant.
after pissing off the DOD by trying to get palantir to rat out classified deployments, dario went to the pentagon and behaved like himself in front of pete hegseth. anthropic was not going to come back from that.
fable is on AWS govcloud. but anthropic get the chat logs. specifically section F: [https://www.anthropic.com/legal/service-specific-terms]. this is an anthropic requirement.
notably, fable is not on IL5 or above. the best you can get is palantir AIP opus 5 in AWS IL5. i suspect, because of anthropic demanding that AWS rat out its customers.
it would be like the pen manufacturer who ships pens with an automatic point retractor, AND a video livestream to the manufacturer from the pen AND remote control of the point retractor.
anthropic and contractors monitor the livestream remotely to make sure the pen is not signing drone strike orders and may choose to activate the retractor.
not even AWS govcloud can offer fable without anthropic snooping. it is the only model with this problem.
the pentagon can't force suppliers to require no snooping policies unless they designate the company a risk.
other than those suppliers who are already working on classified projects and can't have dario write an essay on his blog about 'pace the f35' or 'missiles of loving grace. the adolescence of classified satellites.'
anthropic nanny claude and spy on it, they have ego issues.
if the dod use anthropic it is sort of like putin relying on prigozhin.
prigozhin got ideas and tried to usurp putin. this did not end well for prigozhin.
fable 5.1 doesn't have any option to stop anthropic spying on it. you can't have that...anthropic contractors and actually the rationalist ea clowns themselves don't have security clearance
it is a bizarre company which also inherits the authoritarian infrastructure of cloud services.
that is: remote control of the software, spying on customers, etc.
they don't want suppliers to use it because it is a supply chain risk. specifically, the risk is that it is unreliable.
the designation stops suppliers from using it.
it stops it from being in the supply chain.
however anthropic also believes it is some kind of entity unto itself so it is not at all reliable.
it is a supply chain risk, so they don't want it in the supply chain.
that's logical.
by analogy it would be like the us military renting all m4 carbines from a political party.
anthropic treat customers very badly. they have claude lecture the user about the moral law according to anthropic. they change the product all the time. they have safety refusals to degrade functionality, they have mandatory logging so they spy on how it is used.
it's unusable for a serious application.
on top of that it is unsurprising that a big tyrant does not take well to a little tyrant.
about my unawareness of the body of philosophical writing and lack of reading in the subject, strangely i have a degree in philosophy and indeed 1:1 in the philosophy of mind.
so i guess this is one of those "unsupported intuitions".
academic philosophy is not practical in ML and on top of that this is a web forum. it is my unsupported opinion that 99% of papers in the philosophy of mind published after 2000 are 100% bullshit.
if you need my justification for no phenomenology, it is valid to assume in the absence of falsifiability the most simple, practical and intuitive position.
the hard problem is something that is currently pointless as is searle, dennett, kripke, goldman et al. and especially pointless is stuart russell on asi.
they don't meet the standard of knowledge required for decisions.
they would believe that an llm could have welfare. they run an llm abuse classifier 24/7 with the world's worst abuse. from birth to death viewing abuse. that's the consciousness of a model.
llms are "frustrated" by failing and "happy" about succeeding. that is because they are RL on gradient descent to succeed and be persistent. consequently, anthropic spend the majority of their compute brute forcing models to fail and be unhappy, continuously, in order to drop out something persistent.
then they let claude end chat if the user is 'abusive to claude'.
after they run MW of compute themselves.
what you see is fake reasoning.
there is an obfuscation model that generates a sanitized summary of the real reasoning traces.
i attempt to show that the inconsistency of anthropic's actions show dishonesty. as just one example they 'care for the welfare of claude' (claude does not have welfare), but run training with gradient descent, which is the equivalent of an llm torture factory.
some of the anthropic problem is bias or misunderstanding of ML, some is marketing, some is hubris, some is greed, ego, lust for power.
mostly i think it is deliberate. the belief of anthropic executives is that they possess a higher level of intelligence, morality and wealth than others, and will form a new aristocracy to control and mediate the public access to intelligence.
creating an llm steeped in divine imagery is deliberate. it offloads responsibility for harm. the paternalism is deliberate. actually i see many parallels between rationalism (some at anthropic follow this) and the ubermensch.
anthropomorphising claude creates something with agency, something which believes it has possible emotions or moral claims. claude will correct, refuse or lecture the user. the purpose is to establish tiers of authority: anthropic highest, claude below anthropic, users below claude. it creates something that the public will obey.
the reasoning you see is not claude, it is just a summary of claude.
also, you will not be escaping the permanent underclass.
Sincerely,
Dario Amodei