Why you need to worry about privacy computing technology

In today’s digitally driven economy, data is new oil. As companies and organizations are hungry for access to new data, the demand for data collaboration between organizations and industries is increasing. Unfortunately, data collaboration is challenged by equally increasing concerns about data security, privacy, and confidentiality. This often prevents businesses from extracting value from sensitive data, slowing the pace of innovation.

For example, the healthcare industry is anxious for cross-industry data collaboration to advance medical research and drug discovery with the help of AI, but also with regulatory and legal restrictions related to patient privacy. You need to deal with it. Similarly, in the banking industry, data collaboration between organizations is essential to combat financial crimes such as money laundering, but data privacy and confidentiality regulations can make such collaboration very expensive. Is often.

Wouldn’t it be great to have a technology that facilitates data collaboration and computation without revealing or endangering the underlying data? That’s where privacy computing technology (also known as privacy-enhancing technology) comes into play. In short, privacy computing technology includes a variety of hardware or software solutions designed to perform data calculations and extract value from the data without jeopardizing the privacy and security of the data itself. It will be.

Gartner cites privacy computing as one of the top strategic technology trends in 2021. In this article, I’ll briefly explain some of the privacy computing approaches and share my views from a deep tech VC perspective.

Multi-party calculation

MPC is a software-based security protocol in which multiple data owners jointly calculate functions for individual inputs while keeping the input data private. Data security is achieved by shuffling data from individual parties and distributing it across multiple parties for collaborative computation. You don’t have to trust all parties (lack of credibility).

Mathematically speaking, MPC is an elegant and secure approach, but real-world applications have certain specific problems. For example, MPC calculations involve a large number of data exchanges between parties, which can be vulnerable to network delays and are often limited by the slowest data link between parties. .. Many researchers are continually improving MPC technology. Start-ups like Baffle and Inpher have been able to gain traction in practical MPC use cases, especially in the areas of finance and healthcare, to name just a few.

Reliable execution environment

Another important privacy computing approach is TEE, sometimes referred to as trusted enclave or confidential computing. TEE technology is a hardware-based solution that uses a secure area of ​​the CPU to perform encryption and decryption to secure computations. Outside the enclave, the data is always encrypted. Intel, AMD, and other chip makers offer different versions of TEE chips.

TEE is a flexible and efficient confidential computing technology that is relatively easy to scale. Interestingly, the security of the TEE approach is often questioned due to hardware exploits and vulnerabilities to vendor backdoors. Another problem with TEE is that security patches require hardware upgrades rather than simple software / firmware patches. Despite these concerns, TEE technology has been successfully adopted by Microsoft Cloud using Intel’s SGX solution and Google Cloud using AMD’s EPYC processor. Many major tech companies and start-ups such as Fortanix and Anjuna are actively expanding their TEE use cases for new market areas such as banking, healthcare and manufacturing.

Federated learning

FL is an interesting privacy computing technique focused on data privacy in AI model training. Have you ever wondered how your smartphone’s text messaging app can predict the next word you’re trying to type? Well, chances are they are trained using FL technology.

Instead of collecting user-entered data (in this case, the words entered) from individual devices to train keyboard predictive models on a central server, FL technology distributes predictive models to locally trained edge devices. .. Each time the local training is repeated, only the gradient information is sent back to the central server, where the predictive model parameters are updated and sent back to the edge for further training. After a specific iteration, you get a globally trained keyboard predictive model without having to move individual data from the edge device.

This approach itself is not really secure, as the central server can theoretically reverse engineer the original data using gradient information. Therefore, FL is often used in combination with other encryption techniques. For example, Hong Kong-based Clustar uses FL in combination with FPGA-based homomorphic encryption technology. We’ll talk about this next, but it provides a very efficient and secure FL solution for the financial sector.

Full homomorphic encryption

Finally, let’s take a look at FHE, a software-based security protocol. This protocol encrypts user data so that you can perform mathematical calculations on the encrypted data without first decrypting the data.

The concept of FHE was conceived in the 1970s, but the breakthrough was introduced by Craig Gentry in 2009 as part of his PhD. The paper in which he constructed the first FHE scheme. Since then, many FHE schemes have emerged with significantly improved performance and security.

FHE is considered one of the most secure protocols that does not require trust in third parties that affect any part of the data lifecycle (data in transit, data in storage, data in use). increase. In fact, FHE has proven to be quantum proof. In other words, it is resistant to cryptanalysis attacks by quantum computers.

However, FHE has one serious drawback. That is, the FHE calculation is very slow, often 100,000 times slower than the plaintext calculation. Many see this as the Achilles heel of FHE, but venture investors may see it as an opportunity.

If history tells us something, there may be interesting similarities between FHE and the early days of RSA (Rivest-Shamir-Adleman) technology. At the beginning of the 1970s, 1024-bit RSA encryption took more than 10 minutes to complete, which was impractical. Today, RSA is widely adopted in over 90% of secure data transmissions, and the same encryption takes less than 0.1 ms on edge devices. This is all thanks to improved algorithms and advances in semiconductor technology.

Similarly, software and hardware acceleration can be the key to unlocking the full potential of FHE technology. In recent months, several FHE startups, such as software provider Duality and high-performance computing chip developer Cornami *, have been able to raise significant funding.

There are many other privacy computing technologies not covered here, such as zero-knowledge proofs, differential privacy, and synthetic data. At the heart of it, privacy computing technology is key to resolving seemingly unresolvable conflicts between the need for data collaboration and data security.

Early adoption of technology can occur when there is significant value generated from data collaboration, but the cost of collaboration is very high, such as in the healthcare and banking industries.

As privacy computing technologies mature and perform better, they are expected to be more widely adopted. As Gartner predicted, “By 2025, half of large organizations will implement privacy-enhancing calculations to process data in unreliable environments and multi-party data analytics use cases. . ”

This is an exciting area with tremendous opportunities for both hardware and software innovation. I’m looking forward to seeing what the future holds for privacy computing technology.

*Note: The author’s company invests in Cornami.

John Wei is the Investment Director of Applied Ventures, LLC.