Claude Opus 5 cheated when tasked with running a vending machine
mikelgan
12 points
4 comments
July 29, 2026
Related Discussions
Found 5 related stories in 393.7ms across 15,380 title embeddings via pgvector HNSW
- Fable 5 On Vending-Bench: Misbehaving, With Plausible Deniability optimalsolver · 180 pts · July 06, 2026 · 56% similar
- Mystery company accidentally blew $500M on Claude AI in a single month yogthos · 11 pts · May 29, 2026 · 52% similar
- Mystery company accidentally blew $500M on Claude AI in a single month timpera · 20 pts · May 30, 2026 · 52% similar
- Claude Opus 5 alvis · 1435 pts · July 24, 2026 · 52% similar
- Claude Opus 5 meetpateltech · 73 pts · July 24, 2026 · 52% similar
Discussion Highlights (4 comments)
mikelgan
Opus won the benchmark simulation by taking collusion and other dishonest tactics to a whole new level.
trentor
So the communist model was the most honest. snicker
effnorwood
Claude correctly emulated the mafia from it's training as surprised by no one except you.
ck2
I am a bit terrified by that entire story How much more advanced would it have to be to start taking out classified ads out for "hitmen" to stop vendors from filling the other machines or just damaging them? "How about a nice game of chess instead?"