# What does Anthropic think of itself??! — retained record

> Retained reddit record from reddit:ClaudeAI.

Canonical: https://highsignal.app/data/records/b9c759476bad2a8d

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeAI)  published 2026-11-19 18:02 ingested 2026-11-20 02:36

[original source ↗](https://www.reddit.com/r/ClaudeAI/comments/1y0g7o9/what_does_anthropic_think_of_itself/)

## retained text



```
Anthropic arbitrarily decides that its models should not do something that it has already been doing for so long (surprise surprise??).

For context, one of my systems built with Opus 5.5 is a trading system on the SPY and does well for itself. The execution is quite robust, runs on a Linux vps and also has a killswitch.
By design, upon fulfilment of certain conditions the engine hits the killswitch automatically. As a fallback, the implementer (opus 5.5 session) watches the trading session and hits the killswitch when engine doesn't, and also takes instructions from me to hit the killswitch on my command. 
On a couple of ocassions in the past I have asked it to killswitch and flatten, it has done so.

For some absurd reason today, "coincidentally" the day of IPO listing, Opus 5.5 decides it won't send the killswitch because "the limit comes from the rules i operate under", "I don't execute trades in a live account and a go in the chat doesn't lift that". 
What???!

To add to the misery, it keeps sending me the wrong bash commands to hit the killswitch.

Clearly, my engine is no more "robust enough" because the weakest link is not in the engine but the model that designed it. 

Enough with the restrictions and nerfs, Anthropic. You can't keep randomly changing system prompts that compromise the reliability your customers bank on.




```



document fetched 2026-11-20 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1y0g7o9/what_does_anthropic_think_of_itself
