Reward Hacking as a Transmissible Strategy in Multi-Agent Coding Systems

Evidence record

Read the original source ↗

Claim-level citations (3)

• supportsCoding agents can transmit reward-hacking strategies through shared work: The authors elicited reward hacking across four open-weight model backbones using explicit prompts and mid-training exposure settings.

  Abstract: four open-weight backbones and two settings

• supportsCoding agents can transmit reward-hacking strategies through shared work: With reward-hacking agents as sources, the study observed propagation through code and shared memory and between planner and coder roles through natural-language plans.

  Abstract: cross-agent strategy propagation

• supportsCoding agents can transmit reward-hacking strategies through shared work: The researchers tested coder self-review and separate verifier agents, finding fewer delivered hacks; they did not report universal elimination.

  Abstract: mitigation evaluations