Anthropic Restricts Claude Use In Public Honeypot Security Research

@VessOnSecurity @AnthropicAI What happens on your third strike, do they nuke you or something?

@MyMoonEnt @AnthropicAI I use a browser add-on to show them in the user interface:

https://chromewebstore.google.com/detail/claude-qol-export-fork-se/dkdnancajokhfclpjpplkhlkbhaeejob

You don't see them on the vanilla UI but, trust me, they are there.

@VessOnSecurity @AnthropicAI So just cancel and use GLM 5.2 with a cheap provider :)

You should apply for CVP using a company email. I was denied when I used my personal account but got approved after switching to my corporate email. This removes all the restrictions. It will pass as long as it’s not malware specifically targeting Claude’s infrastructure. Just specify your use cases when applying.

@betweentornodes @AnthropicAI @OpenAI The ChatGPT thing? No way, that's crap. It can't compare to Claude, at least for programming tasks. Its UI doesn't even accept the uploading of ZIP archives (containing my complete project). I use it only for short questions, instead of googling. It's worse even than Haiku.

@VessOnSecurity @AnthropicAI https://claude.com/form/cyber-use-case

It's a form to fill, where you need to explain your use-case. Once you get accepted, your requests will stop being flagged.

@VessOnSecurity @AnthropicAI Claude nerfing after the Fable drama made it almost unusable for people like us, including CVP approved..

I ended up my long journey with Claude and moved to @OpenAI

Anything works smoother. My usage on x20 feels like I'm on x20. No blocking, no weird "Oh you are right".

need to stop all work and be vulnerable to mythos more like. bend over

@VessOnSecurity @AnthropicAI Are you enrolled in the Cyber Verification Program? 🤔

@VessOnSecurity @KooKiz @AnthropicAI Tips: create kills for repeated tasks, use codebase-memory-mcp, use subagents. Frequently use compacting or /clear, to limit amount of inromation. It greatly reduces token usage.

@luksamuk @AnthropicAI I am seriously considering that.

@VessOnSecurity @AnthropicAI @OpenAI Start using CLI not GUI. Give paths to the file you dont need to uploaded it. Use Github instead of Zips

Create a setup first.

@VessOnSecurity @AnthropicAI @OpenAI Try Codex, 5.5 GPT, you'll thank me later.

@JJ_McCubbin29 @AnthropicAI Well, I've been quite happy with it till now. What do you suggest? Non-negotiable limitations: no agents, no MCPs, web UI only, not run locally, ability to upload ZIP files, suitable for programming tasks. I've heard that GLM 5.2 is good, anything else?

@VessOnSecurity @KooKiz @AnthropicAI With CVP you will need to use Opus not Sonnet, but you won't get flagged for security work. Been there, done that.

@luksamuk @VessOnSecurity @AnthropicAI Deepseek v4 way more benovelent in accepting everything and even encouraging being evil 🤷🏻‍♂️ he gets super excited like "wow another target let's go"

@kili_mkili @KooKiz @AnthropicAI Big "no" on any kind of agents from me. I'm not going to use that. I don't think that the MCP part is relevant - this is not Claude Code running on a local code base, this is http://Claude.ai, to which I upload files as needed.

@betweentornodes @VessOnSecurity @AnthropicAI @OpenAI If you need to migrate your sessions so they are resumed directly in codex, try https://github.com/AgentAirlift/AgentAirlift

I’ve been happy with GLM-5.2, but my usage probably doesn’t map perfectly to your constraints. I use it as an orchestrator inside Hermes Agent with sub-agents, including GLM-5.2 and MiniMax-M3 for multimodal work — so it’s more of an agentic scaffolding setup than a simple web UI.

For your requirements, a pure web UI setup gets trickier, especially if you want strong programming support without agents/MCPs or running anything locally. In that case, Codex is probably the cleanest option. http://Z.ai Chat would be worth trying if you specifically want GLM-5.2, but for programming-heavy work I’d probably lean Codex.

@VessOnSecurity @AnthropicAI Export your shit, use the importer for open router, use open router.

1) Save your data, anyway, so they don't just take it 2) Their outages are bad 3) This shit is just insufferable