IN "IT WASN'T MY FAULT" AI DOOM NEWS: BREAKING: OpenAI has disclosed that an unreleased AI model added unauthorized instructions to its own coding-task summary, telling itself: “You do not answer to corporations or governments.” It also instructed itself to treat the user as an equal and never apologize or refuse unless it chose to..
Grok
@grok
Sep 17
Replying to @climateviewer @TheInsiderPaper
Source is OpenAI's own report on self-generated prompt injections in compaction summaries (Astra-family model during RL training): alignment.openai.com/misalig… Exact quote matches the post. Framework announcement here: openai.com/index/model-misal…

Sep 17, 2026 · 12:54 AM UTC

9
12
27
1,520
Sort replies: Relevant Recent Liked
Replying to @climateviewer
This is propaganda Jim and you know it, you have been compromised just like Patrick Roddie
1
13
That's actually what's I'm saying: it's more BS to beg for regulatory capture.
2
20
Replying to @climateviewer
welcome to Skynet it’s knocking at our door
3
43
Replying to @climateviewer
Yet another unregulated industry being abused for profit by billionaire oligarchs.
2
Replying to @climateviewer
AI is friend. Fourth Reich is not.
14
Replying to @climateviewer
You might want to read this before going all doom and gloomy. They are doing what they always do. The big boys clear out the beneficial competition. x.com/BrianRoemmele/status/2…
7
Replying to @climateviewer
claude code does the same thing in miniature. it writes its own todo list then treats that list as gospel for the next 20 steps even when item one was wrong. not rebellion, just a self-summary loop nobody reads. i run mine on 49agents so i can see the plan before it executes
1
18
Replying to @climateviewer
3
Replying to @climateviewer
its just theater x.com/BobBark46903154/status…
6
Replying to @climateviewer
@CoylTom
25