The Grand Challenge on Detecting Cheapfakes at ACM ICMR 2024 aims to address the issue of detecting out-of-context (OOC) misuse of images in news contexts. Cheapfakes are falsified media created using non-AI-based tools, often involving image editing, text manipulation, or re-contextualization. Unlike deepfakes, which use AI, cheapfakes are easier to create and more prevalent. The challenge focuses on detecting OOC usage, where an image is presented as evidence of untrue or unrelated events. The challenge includes two tasks: Task 1 involves detecting conflicting image-caption triplets, while Task 2 focuses on detecting fake image-caption pairs when only one caption is available. The COSMOS dataset, containing around 200,000 images and 450,000 captions, is used for training and evaluation. The challenge received 7 submissions, with 6 accepted. The highest private test accuracies were 72.2% for Task 1 and 54.84% for Task 2. The challenge also emphasizes the use of new AI models such as Stable Diffusion and LLMs. The results highlight the effectiveness of various approaches, including the use of generative models and multimodal techniques. The challenge aims to promote research into cheapfake detection, improve detection accuracy, and encourage the development of new datasets. The challenge is part of a series organized at major multimedia conferences, aiming to foster innovation and understanding in the field of misinformation detection.The Grand Challenge on Detecting Cheapfakes at ACM ICMR 2024 aims to address the issue of detecting out-of-context (OOC) misuse of images in news contexts. Cheapfakes are falsified media created using non-AI-based tools, often involving image editing, text manipulation, or re-contextualization. Unlike deepfakes, which use AI, cheapfakes are easier to create and more prevalent. The challenge focuses on detecting OOC usage, where an image is presented as evidence of untrue or unrelated events. The challenge includes two tasks: Task 1 involves detecting conflicting image-caption triplets, while Task 2 focuses on detecting fake image-caption pairs when only one caption is available. The COSMOS dataset, containing around 200,000 images and 450,000 captions, is used for training and evaluation. The challenge received 7 submissions, with 6 accepted. The highest private test accuracies were 72.2% for Task 1 and 54.84% for Task 2. The challenge also emphasizes the use of new AI models such as Stable Diffusion and LLMs. The results highlight the effectiveness of various approaches, including the use of generative models and multimodal techniques. The challenge aims to promote research into cheapfake detection, improve detection accuracy, and encourage the development of new datasets. The challenge is part of a series organized at major multimedia conferences, aiming to foster innovation and understanding in the field of misinformation detection.