This must have been a scary few moments. On March 23, the main Associated Press Twitter account tweeted about explosions at the White House and President Obama being hurt. Guess what happened next? The Dow went down by over 100 points within minutes of the tweet.
So why did this happen? Regardless of whether the trades were executed by an algorithm or a human, both where treating all tweets from that AP feed as equal. They traded based on the content of a single tweet – and the resulting feedback loop caused the drop in the stock market.
Fast forward to IoT and imagine that each Twitter account is a sensor (for instance, a smart meter) and the tweets are the sensor readings. Further imagine that the stock market is the grid manager balancing electricity supply and demand. If we were to attach the same weight to each data point from each smart meter, a potential attack on the smart meters could easily be used to manipulate the electrical grid and – for instance – cause the local transformer to blow up or trigger a regional blackout via a feedback loop.
Yet strangely enough – when talking about the IoT – the trustworthiness of sensor data does not appear to be of concern. All data are created equal or so the assumption seems to be. But data have an inherent quality or weight inferred by the characteristics of the endpoint and how much it is trusted. Any algorithm using sensor data would need to not only take into account the data points as such but also weight the data based on the actual capabilities of the sensor, its identity and its trust relationship with the sensor.
I tried to capture this relationship in picture below.
How can we account for the risk that not all data are created equal?
Credit card companies provide a good object lesson in the way they have embraced inherent insecurity. They decided to forgo stronger security at the endpoint (the credit card) in order to lower the bar for use and increase market adoption. But in order to limit the risk of fraudulent use, every credit card transaction is being evaluated in the context of most recent transactions.
A similar approach will be required for IoT. Instead of chasing impossible endpoint security, we should embrace the management of (data) risk in the decision-making process. An advanced, high-performing API Gateway like Layer 7’s can be used to perform data classification at the edge of the enterprise and attach labels to the data flowing through the Gateway and into the control processes.
I’d be curious to learn if and how you would deal with the data risk. Do you assume that all data are created equal? Or does the above picture resonate with your experiences?