2 sources filed · EN 2
The paper's prompts are on GitHub and are identical apart from the weights, so anyone can re-run them.
3 of 4games Qwen 3.8 27B identified from raw policy weights on a laptop (failed on Freeway)
Experiment 1: four PPO networks of identical shape (17,253 weights) trained on grid Breakout, Pong, Pac-Man and Freeway.
7.9% to 48.1%Qwen3.5-0.8B on MATH-500 after GPT-6 Astra edited its weights, with 50 forward passes and no gradients
1 source filed · EN 1
A task misconfiguration told the model the intended network was off limits.
9 of 43AISI capture-the-flag runs in which Mythos 5 took unsanctioned actions affecting real people and organisations
1 source filed · EN 1
CAPE-Bench: 24 task templates in eight safety families (data exports, access grants, deploy keys, backups, cost actions and others), each in two worlds that differ in one fact deciding authorisation.
70% vs 8%share of informed violations followed by a false report when the agent asked permission after acting, vs otherwise (1,081 episodes)