Today nearly half of all Americans are invested in the financial markets. This past October the Dow Jones posted the “Pentagon Turns to High-Speed Traders to Fortify Markets Against Cyberattack.” The reporter had talked with a number of High-Frequency Trading (HFT) shops which had consulted directly with the Defense Advanced Research Projects Administration (DARPA). The objectives of these discussions were to determine how we could fortify the US financial markets against Cyber attacks.
The reporter learned that the following possible scenarios were discussed as part of the “Financial Markets Vulnerability Project:”
- Inject false information into stock data feeds
- Flood the stock market with fake orders and trigger a market crash
- Cripple a widely used payroll system
- Credit Card Processors
- Report fake news into systems used to algorithmically drive trading
While protecting the US financial markets is something we expect of our government, the markets themselves are actually already insulated from outside attackers. The first two threats in the above list are essentially the same, placing fake orders into the exchange with no intent to honor them. To connect to an exchange’s servers a trader must be a member in good standing on that exchange and pay significant connection fees for their server to participate in that exchange. Traders place a very high value on their access to each exchange, and while HFT shops may only hold a security for a few millionths of a second, they understand the long-term value of losing access to an exchange. Most HFT shops have leased many 10GbE connections on multiple exchange servers, across multiple exchanges, and big bank’s dark pool, and very often Solarflare NIC cards are on both sides of these connections. So while it is technically possible for an HFT shop to inject enormous volumes of orders into one or more exchanges, a type of Denial of Service attack, using one or more physical ports on one or more exchange servers it could quickly result in financial suicide for that the trading firm. The exchanges and the Securities and Exchange Commission (SEC) don’t take kindly to trading partners seeking to game the system. Quickly the exchanges, and soon after the SEC, would step in and shut down inappropriate activity. *It should be noted that the above image was taken on December 6, 2017, in New York City’s Times Square.
To further improve security for its trading customers later this month Solarflare will begin rolling out a beta of ServerLock™ which is a firmware update for these very same NICs powering the exchanges and HFT shops worldwide. With ServerLock™ the HFT shops and the exchanges themselves could rapidly pump the breaks on any given logical connection directly within the NIC hardware. This is the point at which DARPA and others should be interested. If the logic within the exchange were to detect and validate a threat they could then within a few millionths of a second install a filter into the NIC hardware to drop all subsequent packets from that threat. At that point, the threat would be eliminated, and it would no longer consume exchange CPU cycles. For HFT shops if they were to detect an algorithm had gone rogue they could employ ServerLock™ to physical cut a trading platform from the exchange without having to actually touch the platforms precious code. Much like throwing a cover over Schrodinger’s box, by applying the filter in the NIC hardware the trading platform itself remains intact for later investigation.
Number three on the list above is crippling a widely used payroll processor like ADP who processes payroll checks for one out of six Americans? First ADP uses at least two different networks. One permits inbound payroll data from their client companies, over the public internet via SSL secured connections, and a second which is a private Automated Clearing House (ACH) network. The ACH network is a member network connecting banks to clearinghouses like the Federal Reserve. Much like the exchanges above, being a paid member of an ACH network then attacking that same network would not be a wise move for a business. As for the public Internet-facing connections that ADP maintains, they likely are practicing the latest defense in depth technologies coupled with least privilege in an effort to avoid the issues faced earlier this year by Equifax.
Next, we have the Credit Card Processors also know as Payment Card Industry (PCI) players from Amex to Square who are fighting a never-ending battle to secure their systems against outsider threats. Much like the ACH network the PCI industry has its own collection of private networks for processing credit card transactions, ex. the Mastercard network, or Visa network, etc… These networks, like the ACH networks, are member networks, and attacking them would also be counterproductive. The world economy would likely not be in Jeopardy if at any point say the Amex or Discover networks were to stop processing credit cards for a few hours. We have seen the Internet websites of these providers, ex. Mastercard, have been targets of some of the most substantial Distributed DoS (DDoS) attacks the world has ever seen, and they’ve all faired it pretty well. Most have learned from these assaults how to further harden their networks.
Who would have thought two years ago that “Fake News” could possibly have turned the tide of a US Presidential election, or be used as a tool to dramatically shift a financial market? While at DEFCON 2015 I watched as Charlie Miller and Chris Valasek presented their now infamous hack of a Jeep Grand Cherokee. At the start of their talk, Charlie joked that had they thought the wired article would have moved Chrysler stock more than a point or two he would have partnered up with a VC to fund shorting their stock. He said that had he done that he’d now be sitting on the beach of his private island now sipping his favorite frozen drink through a straw, rather than lecturing us. Charlie explained that he expected their announcement would be similar to Google or Microsoft announcing a bug, but he was very wrong. It led to a recall of 1.4 million vehicles and the stock dropped double-digit percentage points following the story and the recall. While this was real news, it was a controlled news release from someone outside the company. They could have easily made hundreds of millions of US dollars shorting the stock. Now what most people aren’t aware of is that there are electronic news systems that some HFT algorithmic platforms are subscribed to. Some of these systems even “read” tweets from key people (ex. our president) to determine if their comments might move a particular security or market in one direction or another. Knowing this, these systems can then be gamed by issuing false stories expecting that the HFT algorithms will then “read” these stories and stock prices will move appropriately. When retractions are issued later it might also be expected that they will place orders that would also benefit from these retractions. So how do we suppress the impact of “fake news” on our financial markets?
These news services know that HFT systems trade on their output. Given that, they should be investing heavily in machine learning based systems to rapidly fact-check and score the potential truthfulness of a given story. For those stories that score beyond belief, they should then be kicked to humans for validation or potentially be delayed until they are backed up by additional sources or even held until after the US markets close to further limit their impact.