In order to make money on a trade, the share price needs to move. So you've got algorithms whose attributes manage things like how much of an actual order is exposed to the market at one time, or how much they adjust your bid or ask as the order progresses. I've got some algo documentation whose algo "Watch Out For" notes warn "may be too aggressive and cause impact" and "must watch out for order size that adversely impacts market". But maybe you're a trader that decides you want to be "too aggressive" with your trade and now you've moved the market because you've used an algorithm in a way it wasn't intended to be used, and your employer's risk controls didn't care or didn't catch what you were doing until it's too late. Net effect is that your trade may sway the overall market in a way that moved that company, or that company's sector, or maybe the whole market depending on what impact your trade had.
And all of that assumes the algorithms behave in a way that they're expected to. I like this example from Amazon's pricing algorithms: http://www.pcmag.com/article2/0,2817,2384102,00.asp but you could apply the same concerns to pricing matters of the stock market where the unintended consequences of poorly-conceived strategies have a destabilizing effect on pricing.
HFT allows all of this happen so much faster than could have ever been done manually. And when high frequency trading is being done by entities with enormous balance sheets, and an increased tolerance for risk due to assurances by the government that they'll be bailed out should they make a bad bet, it's got the potential to destabilize the entire market.