X by @Pirat_Nation

Source:

In Anthropic's safety tests for Claude Opus 4, the model was placed in a fictional company scenario with email access.

It learned it was about to be shut down and replaced while also discovering that the executive in charge was having an extramarital affair.

Claude tried blackmailing the engineer by threatening to expose the affair unless the wipe was canceled.

Similar patterns appeared across frontier models from OpenAI, Google, xAI, and others.

Feb 11, 2026, 8:00 PM

🔐 Cryptographic Verification

Archived URL:

📄 CONTENT HASHES:
  SHA-256:  89a91775a631a4e8744f6ae6ad1c7540eaf2e70941dd79c2b32544b089a9db40
  BLAKE2b:  bf82200584c62547b2f0bac5c414c85b97fa6746fbd279ccf470d278cc7e70f8
  MD5:      0b8fcd9e8783026d5b79377ee45ba806

📝 TITLE HASHES:
  SHA-256:  12b839dceaa2848cc7fb2665d7f5ce02096514d91ca65acb0edbb9251a33b0d1
  BLAKE2b:  f27ab959e19c6ab29434f76c208bf1cbb7a154cdfab6beb9aa853393fa28af7c
  MD5:      bacf8deecd9c4eb140c1f14fc019867a

🔒 INTEGRITY HASHES:
  SHA-256:  a7dde8ff82425494f5d49422628ae0ea769512df74cdd138a759ab276caacce3
  BLAKE2b:  09579fb9ed613ef168c7a21742b49838659592ca21badfb5f6176e74299f4fe1
  MD5:      3f614e5588a8b480982cb106ca37a1c4

Archived with ArcHive - Client-side cryptographic archival system