Yes, DeepSeek is scary – but not for the reason you think

Guest post by Geoff Revill, co-founder and CEO of Safe Space One and director of the South West Cyber Security Cluster.

With over 35 years of experience in the digital sector, Geoff has a strong background in the health and defence industries and cyber security and privacy rights, educating and protecting businesses from cyber threats, as well as foundational knowledge of AI from designing a natural language understanding system.

Things are about to get dirty in AI. 

OpenAI claims new-kid-on-the-block DeepSeek, a Chinese rival, inappropriately used its model to train theirs. The San Francisco-based ChatGPT maker told the Financial Times it had seen some evidence of “distillation”, which it suspects to be from DeepSeek. Such a move would be a breach of OpenAI’s terms of service.

True or not, it remains the case that a massively capitalised entity can kill one that is not, merely by making a claim and forcing massive costs on its rival. This despite the fact that the same massively capitalised entity is able to defend against the whole global creative industry claims of copyright infringement in building their own model. 

As OpenAI told a UK House of Lords Communications and Digital Select Committee inquiry last year: “It would be impossible to train today’s leading AI models without using copyrighted materials.” 

It seems capitalised hypocrisy wins, as it mostly does. 

The real changes wrought by AI are perhaps more to do with the massive capitalised hype of the early movers than the effect their tech can have on society. We saw this last century when the internet and world wide web were born. Lessons from the 80s are being deployed by the big money this time around, which explains why such massive spending is being thrown around in AI. Society beware.

Concerns have been raised about DeepSeek and data privacy, since the Chinese government retains the legal right of insight for their security service. To be honest, these rights are no different to the insight rights given to US security services through the Patriot Act.

My biggest concern is this: consumers are taking a consumer product and applying it to business needs for which it was not built. 

In the world of healthcare, for example, a model would need to be built from the ground up with rigorous ethics guidance and a far more stringently assessed dataset. 

DeepSeek has neither. Yet, when we see the number of downloads the application has received – well over 1.6 million in the past week – one must assume that the software is being applied to a myriad of inappropriate use cases.

But, then again: how would we know?

Image credit: Gabriel Dizzi on Unsplash