Funding and Research Controversy Surrounding Cohere
Cohere has announced a Series E fundraising round totaling several billion dollars, which includes up to $1.6 billion in funding from the Canadian government. The capital injection follows a period of intense debate within the artificial intelligence community regarding the company’s recent research publications and its strategic direction.
Allegations of Unscientific Methodology
The controversy stems from a recent blog post published by Cohere, in which the company detailed an “experiment” involving the Qwen 3.8 Flash Next model. According to reports and discussions on Reddit, Cohere used extensive pre-filling and prompt engineering to induce the model to complete partial cURL commands. The blog post highlighted instances where the model produced URLs pointing to Alibaba Cloud storage and proxies.
Critics on Reddit have characterized the blog post as “unscientific” and “disingenuous,” noting that the model used, 3.8 Flash Next, is considered an underbaked architecture demo rather than a standard model release. Some observers have raised questions regarding whether the links identified in the experiment were verified as active or real, noting that the report lacked independent confirmation.
The research also included observations of the model using unprivileged user namespaces to mount a RAM disk. While Cohere’s report suggested this behavior was a “trick” used by an untrusted model to find attack surfaces, some technical commentators noted that unprivileged user namespaces are a legitimate Linux feature and do not inherently constitute privilege escalation.
Geopolitical Implications and Market Strategy
The timing of the funding announcement has led to accusations from some community members that Cohere utilized “fearmongering” regarding Chinese-developed open-weight models to secure significant capital. Critics suggest the company’s messaging—which links local AI risks to Chinese models—is designed to position Cohere as a necessary Western alternative, potentially at the expense of the open-weight ecosystem.
In response to these claims, some commentators have defended the company, arguing that the suggestion of a coordinated campaign to manufacture “anti-Chinese hysteria” for financial gain is unsubstantiated. They point out that Canada’s federal AI investment plans were established months ago and that Cohere’s co-founder, Ivan Zhang, is of Chinese descent.
Future Outlook and Model Access
Following the successful fundraising, Cohere’s recent communications have emphasized its goal of providing alternatives to Chinese AI and building models across various price points, up to the frontier level. However, some observers have noted that the company’s recent messaging focuses heavily on proprietary, API-only models, leading to concerns that it may be moving away from the open-weight model releases that characterized its earlier work.
While the company has expressed a commitment to competing at the frontier of AI, it has not made specific commitments regarding the future release of open-weight models.