Asset Managers to Staff: Don’t Touch DeepSeek – Yet

Asset managers are nervous – and intrigued – about the new generative artificial intelligence model from Chinese startup DeepSeek, and are reinforcing their AI-compliance policies with employees. Some industry firms have banned access to the model altogether, while researching whether the technology can make their operations more efficient, consultants told FundFire.
With DeepSeek, there are also broader concerns – among them, the potential for sensitive data ending up in the hands of the Chinese government and bad actors, Jim Lewis, STP Investment Services’ head of technology, said in an email.
DeepSeek also collects keystroke patterns or rhythms, “which can be as uniquely identifying as a fingerprint or facial recognition and used [as] a biometric,” Lewis said. There are also censorship concerns to consider, he added, such as responses to questions about Taiwan and Tiananmen Square, which can lead to biases in the model.
“This isn’t an area where managers should try to be first to market, or have fear of missing out – there is too much risk to outweigh the potential for reward,” he said. “Managers need to be careful of not only having their own private, sensitive information, such as trading or marketing strategies being used against them, they need to be hyper vigilant of their clients’ data being compromised and used against the firm or the individuals.”
Lewis said he was not aware of any of STP’s asset manager clients prohibiting use of DeepSeek among staff.
He said there has been a “lot of talk” among managers looking to generative AI platforms such as DeepSeek to drive efficiencies, particularly around trading. But firms should be wary, he said, because they run the risk of exposing intellectual property and might be susceptible to DeepSeek’s “model using its bias to drive trading to a DeepSeek or [a] Chinese-friendly company, or to a bad investment to damage the manager’s portfolio.”
“Not to mention that DeepSeek could also steal the trading model and put it to work in a cheap, fast robo-trading model, without the manager even knowing and with little recourse if they do,” he said.
Read more on what Jim and other subject matter experts had to say in FundFire here: https://www.fundfire.com/c/4761644/640934/asset_managers_staff_touch_deepseek?referrer_module=emailMorningNews&module_order=1&code=Y25Cb1lXMUFjM1J3YVhNdVkyOXRMQ0F4TnpRNE5EQXhOQ3dnT0RFM05UUTFORGs0