Businesses are increasingly using AI to automate processes to gain efficiencies, be more competitive and avoid disruption in their markets. According to a recent Accenture survey, 82% of executives say that their organizations are using data and algorithms based on that data to drive critical and automated decision-making at unprecedented scale.
But what if the data entering the AI algorithms has been compromised along the way, or the algorithms themselves altered?
Admiral Michael S. Rogers of the U.S. Navy and Director of the NSA has said that data tampering could become the greatest cybersecurity threatorganizations face. Kissinger Associates CEO Jami Miscik agrees that a more damaging brand of cyberattack would be where perpetrators don’t steal the data — but change it! She offered potential examples of companies not knowing where their inventories are, or financial services firms having the balances changed in their accounts and noted that such tactics could wreak havoc for businesses and the broader economy at a time when there is already flagging confidence in the media and the political system.
One prime example comes from the oil industry. If hackers access data sources and attempt to shut down an oil rig for a few days, it causes business disruption and lost productivity. But what if those same hackers covertly alter the data, and the oil company spends months drilling in the wrong places? That’s an enterprise-level existential threat.
In an age of manipulated data, deepfake techniques, and unending data breaches, companies need to know that they are using pristine data in their AI systems. Toxic data will wreak havoc in business systems and cripple an organization’s faith in AI and its customers’ faith in it. Before understanding how and why decisions were made, organizations must be able to stand by the integrity of the data and algorithms used by AI. This might be called Verifiable AI — when an organization can provide immutable proof that the data used by their AI systems is unaltered.
So how do vendors implement Verifiable AI into their products to ensure that their AI algorithms are not handling data that has been tampered with? And how do companies build Verifiable AI into their systems to verify that they are using safe data?
Two leading pharmaceutical companies — Merck and Company, Inc. and Amerisource Bergen — have implemented a fast-track project to more efficiently detect counterfeits and irrefutably document compliance at scale. The project is designed to Track and Tracebillions of items throughout their supply chains. In just a few weeks they were able to ensure:
- Full regulatory compliance around sellable return verifications
- Instant verification of the authenticity of returned items
- No replication of manufacturer data required for wholesalers
- Minimal complexity, maximal security
- Immutable, single source of truth provided to all parties, including regulators
- Interoperable with existing Track and Trace solutions
- Scalable-to-consumer scanning at the point of dispense
With this project, these companies were able to achieve full data verification at multiple stages across their supply chain. By extending the real-time Track and Trace ability, they are now able to add a new “Train” element into the process. The verification that is used to ensure that their supply chain is not handling counterfeit goods can also be used to verify that data used to train AI systems for supply chain automation is also safe. For these companies, Track, Trace and Train is an additional benefit of a chain of custody solution, but for many other companies, this will be their core safeguard of their AI systems.
Data tampering is now the single greatest cybersecurity threat that organizations face. From a simple act of revenge by a disgruntled employee, to corporate espionage, or even a nation-state attack, data tampering is an existential threat that cannot be ignored. While AI automation holds the promise of operational efficiencies, it also has the ability to introduce and replicate toxic data across an enterprise, into business systems and decision-making, and expose customers and partners to those very same risks.
The AI automation revolution is already upon us. But its success hinges on the authenticity of the data that drives the AI, and the ability to verify this authenticity.