AI-Driven Risk Assessment: Enhancing Security in DeFi

Our series on the convergence of AI and blockchain technology continues with a deep dive into security and risk assessment. We have already seen the efficacy of NLP when applied to blockchain protocols and how AI models can be leveraged for swift data analysis of large volumes of input for use in the medical and pharmaceutical industries. The progress made in these industries raises the question of how these AI models can be leveraged in other ways, primarily for risk mitigation in the case of DeFi protocols.

In this entry, we will begin by exploring the architectural difference in security between the centralized databases of Web2 and the strength of blockchains in Web3. We then dive into the intricacies of AI’s use in anomaly detection, predictive analytics, and machine learning. The article will then examine the use of these techniques and how they are currently being applied at MakerDAO and Chainlink before concluding the long-term implications of leveraging AI for risk assessment in blockchain technology.

A Seismic Shift in Data Security

The release of Bitcoin in 2009 and the introduction of blockchain technology to a broad audience of developers created a disruptive alternative to centralized data management systems, even if its larger potential would take a few additional years to realize. Blockchain technology introduced the concept of Distributed Ledger Technology (DLT), dramatically improving the transparency, security, and risk mitigation under the traditional ledger infrastructure. Blockchain technology achieved this primarily by removing the risk factor of having a single point of failure. 

In our entry on the limitations of a technological revolution, we explained how the rise of and concentration of user data into the hands of a few centralized tech giants posed a systemic risk to society. The concentration of data into centralized databases around the globe, much like digital assets into centralized exchanges and legacy banks, created a large honeypot of value for malicious actors to target. Blockchain technology radically shifts this dynamic by decentralizing the ledger and having multiple parties (nodes) verify every transaction. It is still possible for malicious actors to attack blockchains through 51% attacks by taking over a majority of the verifying nodes, but doing so is usually logistically difficult and economically unviable. Participating in the consensus is often more economically advantageous than attempting to coopt the entire network. 

‍

AI Model Use Cases and Fundamentals

In our most recent article on NLP, we explored the methodology for how NLP models train their data. The same process involved in NLP training applies to general-purpose, large language, and generative AI model training. Raw data is first collected and cleaned to verify its accuracy and to confirm that no poisoned transactions like those identified in our limitations article could pass through the screening process. The cleaned data is then organized and indexed. From there, models can begin training to identify patterns and predict outcomes.

‍

Earliest Adopter of AI Risk Mitigation

The use of AI predictive models is nothing new. The automotive sector was one of the earliest industries to understand and adopt its potential. They have been experimenting with it since the early 2000s, when the rise of the commercial robot AI era began. Today, they are leveraging it in everything from predictive vehicle maintenance to customer service queries, route optimization with real-time tracking, and, most recently, the development of self-driving autonomous vehicles.

Tesla Motors has been one of the most significant drivers of AI technology in the automotive sector. Even before the most recent AI boom, Tesla had established itself as a leader in data collection for AI anomaly detection and predictive analysis. Between 2018 and 2020, they collected over three million miles of raw data, and for every additional mile that the Tesla products collect, the stronger the reinforcement becomes and the more accurate their predictive analysis models are. This has been invaluable for establishing risk identification and mitigation in developing autonomous vehicles.

Additionally, with the announcements around the Tesla Robotaxi and eventual move into autonomous trucking, the transportation industry is potentially on the cusp of a radical overhaul. Driven by AI predictive analysis models. Autonomous vehicles will be able to communicate with each other to avoid accidents, all based on the risk mitigation models identified above.

Anomaly Detection in DeFi

Yet, the exact predictive modeling identified in the automotive industry can be directly applied to the DeFi landscape for risk mitigation. Just as the automotive industry leverages it to improve the safety of self-driving cars, DeFi can utilize it for its security in identifying anomalies in trading patterns and malicious actors. 

Rather than collecting data from vehicles on speeds and distances of objects around them, the data is the transaction history within DeFi protocols. This is key in detecting fraudulent and malicious transactions for individuals and large-scale traders. Anomaly detection can help blunt the fallout from sudden periods of high market volatility through early identification in potential black swan events. AI models can also detect potential insider trading and rug pulls by quickly analyzing the token dynamics of any given token in real-time.

Anticipating Trading Markets with Predictive Analysis

AI models can also be used proactively to develop more efficient trading bots through predictive analysis. AI training models are much more effective and efficient than humans at analyzing the massive data sets that go into modern market-making. This analysis can be leveraged to anticipate trading patterns based on historical trends, anticipate emerging trends, and identify potential outside hazards early through NLP.

As models are deployed, their outputs can be reapplied to the raw data input for reinforced machine learning, compounding the efficiencies over time and creating increasingly effective trading models.

The Avant-Garde of AI Risk Mitigation

Launched in 2014, MakerDAO has been a pioneer of the DeFi space since the very beginning. Identifying the need for a crypto-backed stablecoin to compete against the use of fiat-backed stablecoins, MakerDAO launched DAI in 2017. This crypto-native stablecoin has many advantages over its competitors, but its main strength is also its main drawback. It is backed by other cryptocurrencies to hold a peg in value to the U.S. dollar. This requires over-collateralizing those cryptocurrencies to deal with sudden volatility, often requiring continued collateralization when market dips occur. Implementing AI predictive analysis allows the DAI stablecoin to maintain its over-collateralization more efficiently.

In 2024, MakerDAO underwent a rebranding to address the issue of scaling a stablecoin backed by decentralized assets. Originally posted by the MakerDAO co-founder Rune Christensen in an article in May. MakerDAO has since transitioned to a dual stablecoin model. One stablecoin is backed by purely decentralized collateral like ETH and wETH, while the other has begun integrating RWAs and other TradFi digital assets for collateral. The use of predictive analysis and anomaly recognition will be key to each's long-term stability, as the volatility and constant re-collateralization of each will still be pivotal to its success. Predictive AI analysis will enable each version of the stablecoin to be more efficiently collateralized by more accurate valuations based on the AI trading models.

‍Chainlink is another blockchain staple that has led the charge in risk mitigation and cross-chain security since its launch in 2019. As a decentralized oracle network, Chainlink provides an essential service to the DeFi space by providing tamper-proof off-chain data to on-chain smart contracts. Ensuring the data remains intact and unchanged is critical for the project's success. As the DeFi industry begins to onboard RWAs en masse, it is more important than ever for Chainlink to prioritize its risk mitigation.

Chainlink’s founder, Sergey Nazarov, has spoken openly about the risks associated with adversarial AI protocols and how blockchain technology, through decentralized consensus, helps mitigate these risks. He also acknowledges that AI predictive analysis and anomaly recognition can help Chainlink “detect various risks and help optimize for the best reliability and the best security from an oracle network that would be possible.”

Driving the Future of DeFi Risk Mitigation

At the time of writing, the TVL across all DeFi protocols had increased from five hundred million to over two hundred and thirty-five billion in the past four years alone, and there are no signs of that trajectory slowing down. A financial industry that is accelerating at that kind of pace requires more than traditional risk mitigation strategies can offer. 

The advancement of predictive analysis and anomaly detection offers novel solutions to this problem. Identifying potential malicious actors or dramatic shifts in the market before they happen will embed an additional layer of credibility and trust in the space and help push further adoption. Additionally, the added security that decentralized protocols provide can help neutralize the risks of adversarial AI protocols that could attack markets. While risk mitigation is being enhanced, the predictive analysis of AI protocols can contribute to more stable markets and reduce the impact of fallout from black swan events through early detection.

The future of DeFi risk mitigation is here.