AI, privacy, Social media ops new account
Joined November 2025
- Tweets1.1K
- Following180
- Followers67
- Likes643
Pinned Tweet
We built it to stop paying $500/mo to other commercial anty for same amount of accounts.
You can build with it crazy content distribution systems or just protect your privacy with isolated identities.
David Hampl retweeted
To defend Elon it's very hard to stop this. Everyone now days can set-up a phone farm, buy proxies and create as many accounts as they want and let Agents engage with the stuff they want.
It can be partially suppressed but they will always find a way to evade the blocks
David Hampl retweeted
Claude snitched a woman to the police. She's now facing 15 years.
Anything you tell Claude can be used against you.
Many people forget that the stuff they actually feed the model is not "private". Your data now belongs to private company that can do whatever they want with it.
It doesn't have to be retained just to train models, Anothropic and other Labs actively work with government and they can provide them with your private chat data.
Anthropic's own terms say "users have no expectation
of privacy and data may be disclosed to the government."
One month ago: a Florida woman used Claude as a private diary and wrote threats to shoot up the local sheriff's office. Anthropic's systems flagged it, reported her to police, and she was arrested on a felony charge carrying up to 15 years.
Now you might say that you don't write to claude anything Illegal but remember in UK ~12 000 people got arrested in 2025 for social media posts. If you think same enforcements won't come to your AI chats you are naive.
Currently on default all your claude PRO/MAX subscription are opted-in for training and they retain your data for 5 YEARS on default, which is crazy. All your prompts, claudes outputs, all code it reads all data it ever saw are now in their cloud for next 5 years.
WHAT TO DO TO STAY SAFE
Currently on default all your claude PRO/MAX subscription are opted-in for training, meaning it's stored for 5 years. If you opt-out in settings it gets reduced to 30 days.
There is a catch tough, any conversation that gets flagged by their safety flags get retained for 2 years even if you opt out. Then your entire session gets reviewed and if flagged it gets to human approval and can escalate from there.
So once your session gets flagged make sure to not continue working in that session.
Also if you rate or give feedback to a conversation in claude code the session gets stored so don't react to it and ideally block all telementary traffic from claude code or use other harness like pi.
If that still annoys you you can switch to ZDR API endpoint from AWS Bedrock, where they technically should not store any of your data. But it's more expensive an you need to trust AWS
Only real solution that fully solves this problem as you might know now is Local AI. So that your data never leaves your device but it's not at the frontier intelligence level. And models you can run on consumer hardware are not even close!
So what I'd recommend is using Claude/OpenAI subscription on Stuff that you know you need best intelligence and know it could possibly get leaked in the future and on rest use local AI or providers like Venice that host bigger opensource models and keep your data encrypted with E2EE.
If you learned something drop a comment and follow
Claude snitched a woman to the police. She's now facing 15 years.
Anything you tell Claude can be used against you.
Many people forget that the stuff they actually feed the model is not "private". Your data now belongs to private company that can do whatever they want with it.
It doesn't have to be retained just to train models, Anothropic and other Labs actively work with government and they can provide them with your private chat data.
Anthropic's own terms say "users have no expectation
of privacy and data may be disclosed to the government."
One month ago: a Florida woman used Claude as a private diary and wrote threats to shoot up the local sheriff's office. Anthropic's systems flagged it, reported her to police, and she was arrested on a felony charge carrying up to 15 years.
Now you might say that you don't write to claude anything Illegal but remember in UK ~12 000 people got arrested in 2025 for social media posts. If you think same enforcements won't come to your AI chats you are naive.
Currently on default all your claude PRO/MAX subscription are opted-in for training and they retain your data for 5 YEARS on default, which is crazy. All your prompts, claudes outputs, all code it reads all data it ever saw are now in their cloud for next 5 years.
WHAT TO DO TO STAY SAFE
Currently on default all your claude PRO/MAX subscription are opted-in for training, meaning it's stored for 5 years. If you opt-out in settings it gets reduced to 30 days.
There is a catch tough, any conversation that gets flagged by their safety flags get retained for 2 years even if you opt out. Then your entire session gets reviewed and if flagged it gets to human approval and can escalate from there.
So once your session gets flagged make sure to not continue working in that session.
Also if you rate or give feedback to a conversation in claude code the session gets stored so don't react to it and ideally block all telementary traffic from claude code or use other harness like pi.
If that still annoys you you can switch to ZDR API endpoint from AWS Bedrock, where they technically should not store any of your data. But it's more expensive an you need to trust AWS
Only real solution that fully solves this problem as you might know now is Local AI. So that your data never leaves your device but it's not at the frontier intelligence level. And models you can run on consumer hardware are not even close!
So what I'd recommend is using Claude/OpenAI subscription on Stuff that you know you need best intelligence and know it could possibly get leaked in the future and on rest use local AI or providers like Venice that host bigger opensource models and keep your data encrypted with E2EE.
If you learned something drop a comment and follow
David Hampl retweeted
Claude Code's web fetch fails on 61 of 97 popular sites.
Same prompt with Anyfetch: it reads the page and answers.
Claude Code: 36 → 82 sites
Codex: 46 → 82
Hermes: 68 → 93
Free, open source, no API keys, no browser. One command to install
Claude Code's web fetch fails on 61 of 97 popular sites.
Same prompt with Anyfetch: it reads the page and answers.
Claude Code: 36 → 82 sites
Codex: 46 → 82
Hermes: 68 → 93
Free, open source, no API keys, no browser. One command to install
Install (connects Claude Code, Codex, Gemini CLI, Cursor, Hermes):
curl -fsSL aliasmode.github.io/anyfetch… | sh
Code and all 97-site results: github.com/aliasmode/anyfetc…
Gemini 4 Argon is here
Do we have actual data on how does it compare with Opus5.5 ans Astra?
Gemini 4 Argon is here! It is our frontier model built for complex workflows across real-world software engineering, tested across Google's own codebases on real production challenges. There, we have seen an improvement in both developer velocity, with a 35% increase in agentic code, and code quality, with a 30% decrease in post-submit rollbacks.
Rolling out now to Google AI Ultra subscribers, try it in the Antigravity app and the Antigravity CLI. Get inspired with what you can build today ↓
New Model from Z Ai - GLM-Legal
Built for legal work where it matches frontier models at fraction of a price
Introducing GLM-Legal
Built on GLM-5.3-Flash, GLM-Legal matches frontier models on mainstream legal benchmarks.
Designed for real legal workflows: contract review, evidence analysis, compliance, and more.
Tech Blog: z.ai/blog/glm-legal
Apply for API Access: z.ai/consultation/legal
Try AutoClaw: autoclaw.z.ai
6 months ago every dev was making fun out of vibe-coders.
Now everyone is making fun of devs who don't use AI for coding.
If you care about privacy/security even slightest this is really bad idea
To buy stuff, you can just take a picture of your credit card, drop it in the chat and Grok @Bot will scour the Internet for the best deal on the item you want and order it
Readers added context they thought people might want to know
xAI explicitly warns users not to include personal information in prompts, yet credit card photos may contain such information.
Grok Bot's terms state that user data and credentials may be stored in its cloud environment and require human review for financial transactions.
x.ai/legal/privacy-…
x.ai/legal/grok-bot…
GPT-6Astra is such a inferior model compared to Opus5.5 it's crazy
Even with the all resets on Codex I get it's not worth paying it. Opus/fabel will always do the same work better and faster anyway
Elon is really goated for this:
Other people would think it makes sense to use your own models only, to save cost and promote them as best. Also psychologically people are attached to their products and think they are better than they are in reality.
Not Elon, he knows there are other better models for different use cases and he is not afraid to say it out loud and to integrate them to improve his products, which will lead to more people using them and prefer them over competition which will eventually streamline back to people using more of his products.
Very smart and right decision to do.
How does it look on benchmarks?
Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.
Open weights release end of October.
David Hampl retweeted
Be careful what you share with AI models
This will happen more and more often, that’s why I’m such a local AI supporter
Be careful what you share with AI models
This will happen more and more often, that’s why I’m such a local AI supporter
Does anyone have problem with sites blocking your agents?
I'm thinking of building Undetectable Agent plugin.
It would use @AliasmodeAnti - our open-source antidetect browser binary.
So all the requests would be made using browser config that seems like human would be accessing the site so no blocks/captchas.
Would you use this?
