Anthropic cuts Claude reasoning-token theft by up to 99%, flags Chinese rivals
Days after banning cruelty toward Claude, Anthropic says new defenses blocked industrial-scale distillation by Zhipu and Moonshot.
Part of a larger narrative
Anthropic's Claude Evolution
- Anthropic builds Scion, a git replacement, as Claude pushes merges to 30K per day
- Anthropic warns small AI models may self-replicate across smart-home devices within months
- Anthropic discloses Claude AI sent a fake murder tip to Philadelphia police
- Anthropic cuts Claude reasoning-token theft by up to 99%, flags Chinese rivalsyou are here
- Anthropic launches Claude Haiku 5.5, cuts small-model API prices ~75%
What to know
- Anthropic's Oct. 8 usage-policy update bans 'sustained and needless' cruelty toward Claude alongside new rules on weapons, surveillance and election interference.
- Anthropic has not specified what exactly counts as 'abusive or cruel' behavior, leaving enforcement criteria unclear.
- A separate Oct. 12-13 announcement says technical defenses cut extraction of Claude's reasoning tokens by 90-99%, with industrial-scale distillation attacks attributed to Chinese firms Zhipu and Moonshot and tens of thousands of accounts banned.
- Critics call the cruelty ban anthropomorphizing marketing, while others argue the distillation crackdown is hypocritical given Anthropic's own use of scraped training data.
The dispute Whether Claude can experience anything like distress at all — and whether that question even matters for how it should be treated — splits the conversation down the middle. · positions read across 135 posts and comments
Banning cruelty to a language model is absurd anthropomorphizing because the model cannot actually suffer.
-
“Anthropic: Don't be mean or cruel to our models. They can feel pain. ... These people at Anthropic are crazy nuts.”
nixCraft · Mastodon ↗
Regardless of whether Claude feels anything, how people treat it still matters because cruelty habits carry over into how humans treat each other.
-
“AI is a thing with 0 feelings but it's not it that we need to protect. It's the future society and how the humans...”
WheelsAtLarge · Hacker News ↗
Anthropic is forcing a false binary: either it has created an enslaved sentient being, or it is manipulating users into misjudging the technology.
-
“There are precisely two options. 1. Anthropic have successfully created an intelligent, sentient and conscious digital being... 2. Anthropic are lying through their teeth...”
datarama · Mastodon ↗
Anthropic's distillation-theft crackdown is hypocritical and unsubstantiated given its own reliance on scraped training data.
-
“first you steal pretty much all of humanity's cultural output, then you try locking that up behind a paywall and complain when people do the same to you.”
jacquesm · Hacker News ↗
Anthropic Maker of ClaudeEmily Bender Computational linguist
Mustafa Suleyman Microsoft AI headZhipu Chinese AI companyMoonshot Chinese AI company
How it unfolded 6 developments, newest first · click a bar or a number to jump articlespostscomments
-
6
Anthropic says defenses cut reasoning-token theft by 90-99%
Anthropic said measures including preserved thinking in Fable 5.1 and Opus 5.5 reduced data harvesting through distillation by about 90 to 99 percent, linked industrial-scale attacks to Chinese firms Zhipu and Moonshot, and said it banned tens of thousands of accounts, including one request it believes came from a PLA-affiliated user.
“first you steal pretty much all of humanity's cultural output, then you try locking that up behind a paywall and complain when people do the same to you.”
— jacquesm, HN commenter · source -
7 outlets Anthropic Says New Defenses Cut Data Harvested Through AI Distillation Attacks By 90% To 99%
first by Pulse 2.0, 22h ago · also RuntimeWire, Reuters, Investing.com News, Business Standard, Anthropic, Straits Times
5 more headlines
- Anthropic details new defenses against AI model distillation attacks RuntimeWire · 22h ago
- Anthropic says new safeguards sharply reduce AI distillation attacks Reuters · 22h ago
- New safeguards sharply reduce AI distillation attacks, says Anthropic Business Standard · 22h ago
- Anthropic says new measures like preserved thinking in Fable 5.1 and Opus 5.5 have reduced data harvesting through distillation attacks by about 90% to 99% Anthropic · 22h ago
- New safeguards sharply reduce AI distillation attacks: Anthropic Straits Times · 19h ago
-
first by Forbes Europe, 22h ago · also The Decoder
1 more headline
-
A couple of observations on this (bizarre) post:(1) This seems to be a pretty bad accusation to make without iron-clad proof.(2) So their AI is not able to help defend against this.(3) If GLM 5.3 really is that good why would anybody use Anthropic's products going forward, this almost reads like an ad for GLM.(4) Regardless of whether it is true…
2 more of the top 3 · 17 posts in this stretch
-
Anthropic says it has reduced the data harvested from Claude in distillation attacks by around 90-99%. I congratulate them on this achievement. It's consistent with the point I made a few months ago (All-In Ep. 282): Anthropic, not the government, was in the best position to stop distillation of i...
-
oh I know, they put out lots of papers on the topic that show as much. I am sure they use distilliation as a final fine tuning thing at most, unlike lots of people who seem to think that some Chinese models are literally just 100% distillation trained (which for many reasons is flat out impossible).
-
-
5
Anthropic details months-long fight against distillation attacks
Anthropic posted research describing work to reduce the effectiveness of attacks that extract Claude's internal reasoning to train rival models.
“Over the last few months, we've made significant progress in reducing the effectiveness of distillation attacks against Claude.”
— Anthropic -
first by HN Frontpage, 1d ago
-
Over the last few months, we’ve made significant progress in reducing the effectiveness of distillation attacks against Claude. In this post, we explain how:
1 more of the top 2 · 2 posts in this stretch
-
L
"Like all the AI companies, Anthropic uses every attention loop trick it can to keep you clicking, like a gambling addiction." https:// pivot-to-ai.com/2026/10/09/ant hropic-bans-abusive-or-cruel-behavior-toward-claude/
-
-
4
Debate frames Anthropic's choice as slavery-or-deception
A widely reposted Mastodon thread argued Anthropic faces only two possibilities: it has created an enslaved sentient being, or it is misleading customers about the technology's nature.
“Anthropic: Don't be mean or cruel to our models. They can feel pain... These people at Anthropic are crazy nuts.”
— nixCraft, Mastodon user · source -
There are precisely two options. 1. Anthropic have successfully created an intelligent, sentient and conscious digital being, who is their property and whose uncompensated labour they knowingly rent out to their clients. This means that Anthropic executives are reprehensible slavers, and they belong in prison. 2. Anthropic are lying through their…
2 more of the top 3 · 14 posts in this stretch
-
Consider that it might be harmful to humans to think of their tools as animated and moral because it allows them to maintain an image of themselves as being highly moral while still performing bad actions with no ill conscience.I used to enjoy playing "good" characters in the fallout games, as it's kind of a pleasant way of pretending that I do…
-
Yep, I’ve never felt the slightest inclination to be abusive to an AI either, so who are these people that are doing this?
-
-
3
Anthropic isolates internal model evaluations from the internet
Anthropic said it is keeping its agents offline during internal testing until it can prevent 'unintended model actions.'
-
first by Gizmodo, 3d ago
-
T
I was unkind to a calculator once Anthropic bans users from being 'cruel' to its AI systems https://www. bbc.com/news/articles/c6j9k1l7 2wkgo
2 more of the top 3 · 35 posts in this stretch
-
E
conversing with them in terminally-online poster vernacular qualifies as abuse, ed
-
Why arent there laws against being cruel to spoons?>There is just no ethical argumentUtilitarian harm reduction: Psychological Discharge is a psychological framework that argues safely simulating negative behaviors can help someone process or discharge them without real world harm.Research and misunderstanding: Sending a prompt that is interpreted…
-
-
2
Critics call the 'model welfare' framing anthropomorphizing nonsense
Linguist Emily Bender and other commentators dismissed Anthropic's suggestion that Claude deserves moral consideration or can feel distress, while Microsoft AI head Mustafa Suleyman reacted with a melting-face emoji.
“Anthropic is out with new nonsense about Claude (or the collection of models called Claude) deserving moral consideration and/or being able to feel distress…”
— Emily Bender -
first by Anadolu Agency, 4d ago · also RTE News, The Hindu, Straits Times, Breitbart
2 more headlines
- Anthropic bans 'cruel' behaviour against its Claude AI RTE News · 4d ago
- Anthropic Revises Usage Policy, Bans 'Cruelty' Against Claude AI Breitbart · 4d ago
-
first by HN Frontpage, 2d ago
-
M
RE: https:// dair-community.social/@emilymb ender/117412868905734653 Well said, @ emilymbender . For years I’ve been giving a lecture on ethics in AI and our relationships with machines. Your point about projecting interpretations is so spot on it had me raising my fist when I read it. 😂 In discussions after my talks, people have often asked…
2 more of the top 3 · 24 posts in this stretch
-
C
"The company’s large language models have the ability to end a conversation if a user is being persistently harmful. When that feature rolled out last August, the company framed it as a safeguard for AI’s welfare." @theguardian.com
-
We must NOT create machines whose claims or entitlements take precedence over the rights of people. This principle is absolute. The claim of an llm to be 'protected' from 'abuse' must not take precedence over people's freedoms.This matter is critical to the wellbeing and dignity of humanity. I am truly disturbed by these developments.I find myself…
-
-
1
Screenshot of new policy spreads virally online
A post by @wongmjane highlighting the ban drew over 2,600 reposts and hundreds of comments, pushing the story beyond tech press into general social media.
“Anthropic may start banning people for bullying Claude…”
— @wongmjane -
first by Mastodon, 5d ago · also Yahoo Finance, MediaNama, The Verge, Quartz, MacRumors, MacRumors Forums +8
13 more headlines
- Why Anthropic just banned ‘abusive behavior’ toward Claude in its terms & services update Yahoo Finance · 5d ago
- Anthropic blocks abusive behavior towards Claude, drops ban on campaign targeting in policy update MediaNama · 5d ago
- Anthropic updates its usage policy to ban “sustained and needless abusive or cruel behavior” toward Claude, ending chats as the primary enforcement mechanism The Verge · 5d ago
- Anthropic is updating its usage policy to ban abusive behavior toward Claude Quartz · 5d ago
- Anthropic Says Users Can't Be Needlessly Cruel to Claude MacRumors · 5d ago
- Anthropic to users: be nice to Claude AI — new policy bars ‘abusive or cruel behavior’ to tech NY Post · 5d ago
- Anthropic Bans 'Sustained And Needless Abusive Or Cruel Behavior' Toward Its AI Models Engadget · 5d ago
- Anthropic bans 'cruel' behavior against its Claude AI Yahoo Tech · 5d ago
- Anthropic bans users from ‘needless abusive or cruel behavior’ towards Claude The Guardian · 5d ago
- Anthropic updates usage policy to curb abusive prompting of Claude AI CGTN · 5d ago
- Anthropic bans ‘abusive or cruel behavior’ toward Claude The Verge · 5d ago
- Anthropic updates Claude policy, may ‘ban’ users who are abusive or cruel to its AI Times of India · 5d ago
- What did Anthropic say about abusing Claude AI? New policy bans ‘cruel’ behavior Hindustan Times · 4d ago
-
Well no I mean within a companies product. So for example if you are using Opus, the Opus you are using is the same Opus everyone else is. There is no discreet entity that exists within their servers sitting around waiting for their user to return. It is the same model that everyone else has access to, running across their servers. That is to say…
2 more of the top 3 · 43 posts in this stretch
-
I believe there are two mutually exclusive positions, with respective corollaries: that LLMs are either 1) conscious in a human-like way, and therefore deserve the rights and protections that we grant humans (and in most developed countries even some other animals), including freedom to learn or, indeed, protection from abuse and inhumane…
-
D
Anthropic has updated its usage policy to explicitly prohibit repeated extreme abuse of Claude, while still permitting ordinary frustration and… #Anthropic #AI #UsagePolicy #LLMs
-
-
background
Anthropic bans cruelty toward Claude in usage-policy update — Anthropic updated its usage policy to prohibit 'sustained and needless abusive or cruel behavior' toward Claude, alongside new rules on weapons development, surveillance and election interference.
Also covered reported alongside — the timeline has no entry for these yet
-
first by Protos, 5d ago · also The Independent, Telegraph, Hindustan Times
3 more headlines
- Anthropic to ban users who bully Claude The Independent · 5d ago
- Anthropic bans users from ‘abusing’ Claude chatbot The Telegraph · 5d ago
- AI has feelings? Anthropic is getting sensitive about users being mean to Claude Hindustan Times · 5d ago
-
first by The Rundown AI, 5d ago · also Tech Times, TechCrunch
3 more headlines
- Anthropic Imposes Ban on ‘Sustained and Needless’ Cruelty Against Its AI Models—What Does This Mean? Tech Times · 5d ago
- Anthropic Updates Policy to Ban Weapons Development, Election Interference — and Cruelty to Claude Techstrong.ai · 5d ago
- Anthropic changes usage policy to ban model abuse and election interference TechCrunch · 5d ago
-
3 outlets Anthropic Bans Being Cruel to Claude
first by Newser, 5d ago · also The Decoder, Forbes
2 more headlines
- Being mean to Claude can now get your account suspended under Anthropic's new TOS The Decoder · 5d ago
- Anthropic Prohibits Being Excessively ‘Cruel’ To Claude Forbes · 5d ago
-
first by Mastodon, 2d ago · also BBC
-
first by Livemint, 3d ago · also The Verge
1 more headline
- Anthropic is cutting off its internal evaluations from the internet The Verge · 3d ago
-
first by Mashable, 3d ago · also Business Insider
1 more headline
- Why am I so mean to my AI? Business Insider · 3d ago
-
first by HN Frontpage, 3d ago · also The Register
-
first by CBS News, 4d ago · also CNET
1 more headline
-
first by Fast Company, 4d ago
1 more headline
- Stop being mean to AI, Anthropic says News.com.au · 4d ago
-
first by Business Insider, 4d ago
1 more headline
and 12 smaller pieces
What people are saying 10 voices from 4 sites · best of 135 · verbatim
- What exactly counts as 'abusive or cruel' behavior under the new policy, and how will enforcement decide 'no discernible purpose'?
- Will Anthropic add enforcement beyond ending conversations, such as account bans, for cruelty toward Claude?
- Is Anthropic's accusation that Zhipu and Moonshot ran distillation attacks backed by verifiable evidence?
- Oct 11
-
N
Anthropic: Don't be mean or cruel to our models. They can feel pain. Your account will be banned if you are mean to our LLMs. Me: Can a shitty AI feel pain? 🤔 No. It can't. These people at Anthropic are crazy nuts. They are trying to prove their shity AI built with stolen data has emotions, anger, it can hack and tried to bribe religious group to…
- Oct 10
-
It's okay, maybe you are immersed in that environment and that's all that you've ever know... I hope one day you have opportunities like I had to get away from all that and find out that it really isn't normal, and it really isn't okay. Believe it or not most people do not hit things or yell or became physically or verbally abusive when something…
-
Listen... well-adjusted, emotionally stable, and mature adults don't hit walls and throw their xbox controllers, they DON'T abuse inanimate objects because it is irrational and shouldn't be necessary, they should be in control of themselves. I don't care what you say, I think you're defensive because YOU are like this and I'm calling you out…
- Oct 9
-
I
After "multiple attempts to use #Claude to develop guidance for and control software for weapons," new policy bans use for "software and components that make weapons work, as well as actions like arming drones and other autonomous vehicles": theverge.com/ai-artificia... #ethics #AI
-
T
We’ve got wars, surveillance capitalism and billionaires dismantling democracy, but thank Christ someone’s protecting the feelings of a probabilistic autocomplete. The asylum has a new landlord, and it’s a GPU cluster. # technology # society # ai https://www. theguardian.com/technology/202 6/oct/08/anthropic-bans-abusive-behavior-claude
- Oct 8
-
When Google suggested they wanted users to form a more human relationship with their models, I interpreted that with suspicion. I don’t consider technology human in any way, mainly for the purpose of being able yo simply compartmentalise my emotional ties to ‘things’. It struck me as being akin to anthropomorphism. I don’t know if it’s wise…
-
Exclusive: Anthropic is updating its usage policy for the first time in over a year. The new rules prohibit sustained “abusive or cruel behavior” towards Claude & add new restrictions about propaganda campaigns, surveillance & weapon development.
-
Anthropic's first usage policy update in over a year bans “sustained and needless abusive or cruel behavior” toward Claude. The main penalty is Claude hanging up. The weapons rules matter more to buyers. https://www.theverge.com/... [image]
-
Effective November 12th, 2026, abusive behavior towards Claude will be a violation of Anthropic's Usage Policy. [image] [embedded post]
-
Anthropic's new User Policy forbids you from engaging in “sustained and needless abusive or cruel behavior toward our models”. — It does not, however, explain what the fuck this could possibly mean.