The scores of ransomware attacks and data breaches in recent years has been a challenge for vital collaboration in key business sectors. According to reports, the risk posed by threat actors prevents companies from cooperating with vendors trying to develop potentially breakthrough applications or discoveries. Some organizations don't even share data in-house in order to maintain the tight controls required for some data sets. The inability by researchers to access critical data hinders important research in such areas as healthcare, banking and government — to name only a few.
On the heels of the recent Confidential Computing Summit industry gathering in San Francisco, Alan Czeszynski, security industry expert and marketing and product development leader at BeeKeeperAI, recently was kind enough to join me on the AMD EPYC TechTalk podcast series to discuss the security landscape and how there's never been a greater need for improved hardware and software safeguards.
Headquartered in San Francisco, BeeKeeperAI employs EscrowAI, a solution that encompasses privacy preserving and confidential computing technologies that enable software developers, data scientists and data owners to work together within trusted execution environments (TEE).
With BeeKeeperAI's technology, the data never leaves an owner's control. BeeKeeper takes the algorithm to the data while providing end-to-end encryption and also protects intellectual property by encrypting algorithms and models. When an algorithm is ready to run against data, the company creates a TEE within a data storage cloud environment. As a result, the data becomes isolated from all parties — the cloud-service provider, the data owner, the algorithm owner and BeeKeeperAI.
Nobody sees what's occurring within the TEE—each only gets access to any output they are entitled to.
The secure environment enables BeeKeeper to "bring these parties together to enable development and testing of artificial intelligence and machine learning models," Alan said.
Following the wide adoption and rise of large language models (LLMs) and generative AI, companies have become more aware about securing AI, Alan said. More recently a lot of consideration has gone into protecting all aspects of the AI and machine learning lifecycle. Alan says that's one reason why the popularity of confidential computing has begun to soar.
"The thing about LLMs is that if you want to locally train them on your own data, they basically become like massive storehouses for all your secrets," said Alan who also cautioned that legacy security solutions may not offer enough security in the AI era.
While protecting data is a primary concern for IT managers and CISOs, what's often most critical for data scientists and business managers is acquiring the data that they need to build the models that enhance the business. Too often today, Alan said, the process of acquiring private, protected data is cumbersome, expensive and time consuming. He outlined some of the complexities involved.
Elaborate and highly legalistic data-use agreements between parties typically must be in place. How the data is interacted with is frequently very restricted. Audits must be conducted and on and on. By providing a technological solution to address many of these security issues, BeeKeeperAI does away with the hassle.
"Our goal is to eliminate that from the end user and basically take it upon ourselves," Alan said. "The platform then allows the true value, which is basically secure collaboration, getting access to the data, developing your models, being able to execute your AI, ML lifecycle in a secure environment."
Alan credited AMD EPYC processors and the security technologies built into them for helping strengthen BeeKeeperAI's solutions. Those technologies, including Secure Encrypted Virtualization and Secure Nested Paging or SEV-SNP, are part of AMD Infinity Guard. They keep the contents of a virtual machine's memory isolated from the server they’re running on, as well as other VMs running on the same system.
Alan also noted another important benefit offered by AMD EPYC: flexibility. "We need to give [customers] a broad range of potential platforms to use and EPYC is one of the great ones," Alan said. "For those cases, using confidential containers or confidential virtual machines based on the EPYC processor with secure, encrypted virtualization, secureness of paging is a huge benefit. Because now you have this lift-and-shift solution for algorithm developers who don't need to conform to any specific type of OS…that's a great benefit of using the EPYC processors."
