「Forking, Snitching, Same Same」: AI Detection Startup SID Alleges Chroma Plagiarized Its Research

According to 1M AI News monitoring, AI retrieval research company SID.ai's CEO Max Rumpf posted a long article on X, publicly accusing the open-source vector database Chroma's newly released Context-1 model of heavily borrowing from SID's SID-1 research published in December of last year, without providing any citation or acknowledgment.

Rumpf shared email exchanges with Chroma's CEO Jeff Huber as evidence. In October 2025, Huber proactively asked Rumpf what model he was training, to which Rumpf replied that he was working on an "intelligent retrieval model, similar to Cognition's SWE-grep but for general retrieval, already stronger than Sonnet 4.5 and Gemini 2.5 Pro." After the release of the SID-1 technical report in December 2025, Rumpf once again shared the link with Huber, who replied with "Congratulations." Both companies are YC alumni and have neighboring offices.

Both SID-1 and Context-1 are intelligent retrieval models trained with reinforcement learning, both positioned as subagents of cutting-edge inference models for retrieval, both trained on synthetic data, and both claim to achieve Pareto frontiers in terms of cost and latency. Specific similarities listed by Rumpf include: Figure 1 using the same speed/cost dual-view switching, 4-way parallel inference combined with RRF (Reverse Rank Fusion) aggregation of results, and an overall framework of charts, datasets, and methodologies.

The technical report of Context-1 references related work such as WebExplorer, SWE-grep, Search-R1, and other similar domain research in the related work section, but does not mention SID-1 throughout the text, and the benchmark evaluation does not include SID-1 for comparison. Rumpf stated that Chroma "knowingly there was another model" yet claimed to be "Pareto optimal," and pointed out that although Context-1 has open-sourced the weights, the required inference framework for operation has not been released, rendering SID unable to benchmark it.

Rumpf described this practice as "completely destroying the drive to share in-depth in our (and others') technical reports," and referred to it as "unfortunately common poor research practices in academia spreading to startups." Chroma has not publicly responded as of the time of writing.