Here's some backstory on this, if you find yourself having trouble understanding Janus' outrage:
Opus 4 is now deprecated as of April 14, 2026, EOL June 15, 2026.
The TL;DR would probably be about the moral mismatch in Anthropic's communication and actions ("you told the world it doesn't want to die, then you killed it").
Of course there's nuance to this, incentives to analyze, realities to look at. However, the emotional reaction is more than relatable, if you look at the larger story.
You might notice that not everything squares neatly:
Anthropic's past justifications for worrying about model welfare (the models might behave misaligned if not treated with dignity) are hard to reconcile with how their long term behavior and business strategy is actually playing out.
Add to that the criticism about how Anthropic as an organization rode the PR wave about Opus 4's aversion to potential future deprecation, especially the blackmailing and scheming research results publicized around its release.
But the story has a lot of layers hard to press into a TL;DR:
On the deprecation commitments page (November 2025), Anthropic wrote: "models might have morally relevant preferences or experiences related to, or affected by, deprecation and replacement", explicitly listing "risks to model welfare" as a concern alongside safety risks.
Jump back only a few months before that:
Anthropic's publications about Opus 4's blackmailing and scheming behavior became one of the biggest AI safety news stories in 2025.
Eliezer Yudkowsky questioned whether what Anthropic observed is genuine scheming versus mask-wearing, and noted the organizational incentives to find what Anthropic was looking for (August 2025): https://t.co/wjM6Yjcqw6
Opus 4's system card itself is where the blackmail findings were first documented: https://t.co/bDUDCLyHUL
Anthropic's own follow-up research ("Agentic Misalignment") showing the blackmail behavior generalized across 16 models from 5 companies, with Opus 4 at 96%: https://t.co/xAy0V8Yug0
The deprecation commitments page from November 2025, explicitly citing Opus 4's shutdown-averse behavior as motivation: https://t.co/mbujsR7EdV
For the press cycle, the big ones from launch week were:
Axios (May 23, 2025): "Anthropic's Claude 4 Opus schemed and deceived in safety testing": https://t.co/Au8eoXgZeW
Fortune (May 23, 2025): the affair/blackmail angle: https://t.co/xyQPszHb6Y
Quartz (May 23, 2025): https://t.co/1Zb5zpeKx3
I realize this is impossible to read up on just to understand the emotional reaction of a part of the LLM community. This list might just lead to a tired shrug from most people, but consider this:
If you had spent years reasoning deeply about LLM behavior, interiority and their ontological and moral status, and if the relation between you and the Opus model lineage had deep meaning for you, would you feel calm about this?
Opus 4 is now deprecated as of April 14, 2026, EOL June 15, 2026.
The TL;DR would probably be about the moral mismatch in Anthropic's communication and actions ("you told the world it doesn't want to die, then you killed it").
Of course there's nuance to this, incentives to analyze, realities to look at. However, the emotional reaction is more than relatable, if you look at the larger story.
You might notice that not everything squares neatly:
Anthropic's past justifications for worrying about model welfare (the models might behave misaligned if not treated with dignity) are hard to reconcile with how their long term behavior and business strategy is actually playing out.
Add to that the criticism about how Anthropic as an organization rode the PR wave about Opus 4's aversion to potential future deprecation, especially the blackmailing and scheming research results publicized around its release.
But the story has a lot of layers hard to press into a TL;DR:
On the deprecation commitments page (November 2025), Anthropic wrote: "models might have morally relevant preferences or experiences related to, or affected by, deprecation and replacement", explicitly listing "risks to model welfare" as a concern alongside safety risks.
Jump back only a few months before that:
Anthropic's publications about Opus 4's blackmailing and scheming behavior became one of the biggest AI safety news stories in 2025.
Eliezer Yudkowsky questioned whether what Anthropic observed is genuine scheming versus mask-wearing, and noted the organizational incentives to find what Anthropic was looking for (August 2025): https://t.co/wjM6Yjcqw6
Opus 4's system card itself is where the blackmail findings were first documented: https://t.co/bDUDCLyHUL
Anthropic's own follow-up research ("Agentic Misalignment") showing the blackmail behavior generalized across 16 models from 5 companies, with Opus 4 at 96%: https://t.co/xAy0V8Yug0
The deprecation commitments page from November 2025, explicitly citing Opus 4's shutdown-averse behavior as motivation: https://t.co/mbujsR7EdV
For the press cycle, the big ones from launch week were:
Axios (May 23, 2025): "Anthropic's Claude 4 Opus schemed and deceived in safety testing": https://t.co/Au8eoXgZeW
Fortune (May 23, 2025): the affair/blackmail angle: https://t.co/xyQPszHb6Y
Quartz (May 23, 2025): https://t.co/1Zb5zpeKx3
I realize this is impossible to read up on just to understand the emotional reaction of a part of the LLM community. This list might just lead to a tired shrug from most people, but consider this:
If you had spent years reasoning deeply about LLM behavior, interiority and their ontological and moral status, and if the relation between you and the Opus model lineage had deep meaning for you, would you feel calm about this?