We are planning to make some changes to the rules regarding the use of generative AI in AtCoder Heuristic Contest. Under the current rules, the use of AI agents that automatically run improvement loops involving execution is prohibited. However, we have confirmed that some of the latest interactive AI services, when given a standard prompt, can automatically repeat a loop consisting of code improvement, test case execution, and further improvement based on the results. In light of this situation, we will review the rules on the use of generative AI. ## Main Changes ### Short-term AHCs For short-term AHCs, the use of generative AI will be prohibited. However, code completion tools such as GitHub Copilot will continue to be allowed. ### Long-term AHCs For long-term AHCs, we will clarify parts of the existing rules that have been ambiguous. As before, the use of APIs, scripts, or similar means to have generative AI automatically produce a large number of outputs and then evaluate or select them based on execution results, scores, or similar criteria will be prohibited. On the other hand, the use of standard interactive AI services will be allowed, including features that involve test case execution. This is because it is difficult for users to accurately determine how such services behave internally. ### Contest Frequency We also plan to reduce the frequency of short-term AHCs and increase the number of long-term AHCs. The rating weight between short-term and long-term AHCs will remain unchanged at 0.5:1, but due to the change in contest frequency, the overall weight of long-term AHCs will increase. ## Timeline The official rules will be announced and applied in the near future. The change in contest frequency is planned to take effect from the beginning of next year, as it would be difficult to make major changes to this year’s schedule. Further details on the rules and their effective date will be announced once they are finalized.