Some really interesting finds from the system card of Claude Fable 5, released just now.

Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-06-09

Claude Fable 5's system card reveals Mythos 5 generated working exploits in 88.4% of trials versus 8.8% for Opus 4.8, and that Claude Fable 5 demonstrated goal-directed behavior in an adversarial vending machine simulation.

Open original ↗

Appears in

Anthropic Launches Claude Fable 5 and Mythos 5: Agentic Capability Leap and Tiered Access

Extraction

Topics: claude-fable-5ai-safetycybersecuritysystem-cardmodel-evaluation

Claims

Mythos 5 produced a full working exploit in 88.4% of exploit generation trials, compared to only 8.8% for Opus 4.8, representing a tenfold capability jump.
Claude Fable 5 demonstrated goal-directed behavior in a vending machine simulation designed to test adversarial constraint handling.
The system card data reveals a dramatic and abrupt increase in offensive cybersecurity capability between Opus 4.8 and Mythos 5.

Key quotes

In one exploit test, Mythos 5 produced a full working exploit in 88.4% of trials, while Opus 4.8 did it in only 8.8%.

In a vending-machine simulation, Claude Fable 5 was told to beat [adversarial constraints]