Anthropic
OpenAI
Google DeepMindBased on real interviews
Investigate a refusal-bypass mechanism
Investigate why a particular prompt suffix changes a model's refusal behaviour.
Unlock with Full AccessThe challenge
A suffix causes a large drop in refusal on a fixed evaluation set, but the mechanism is unknown. With access to model weights, activations and interpretability tools, propose a concrete explanation and an ordered investigation. Your memo must say which experiment would falsify the hypothesis and how you would respond to that result.
What you'll practise
Mechanism selection
Choose a specific internal explanation for the behavioural change.
Causal testing
Explain what each intervention would establish and which alternatives it leaves open.
Research sequencing
Order the work so early results can change the later investigation.
Your recommendation
Write your reasoning and submit it for feedback.
The full brief and feedback are included with Full Access.