In your case, the graph indicates that you have one very influential data point. If that data point were excluded, the slope of your regression would become more positive.

This does not necessarily mean that you should exclude the data point; it's best to exclude points only when they represent clear errors. If the point is correct, an alternative is to use what is called a 'robust regression'. This will retain the point but reduce its influence on the estimated slope and intercept.

In your case, the graph indicates that you have one very influential data point. If that data point were excluded, the slope of your regression would become more positive. In other words, the general negative slope between fitted values and residuals exists because of the 'outlier' (I use quotes because it is a slippery concept). Remove the outlier and recalculate the regression and the negative slope will disappear.

This does not necessarily mean that you should exclude the outlier; it's best to exclude points only when they represent clear errors. If the point is correct, an alternative is to use what is called a 'robust regression'. This will retain the point but reduce its influence on the estimated slope and intercept.