The inner workings of AI search are more complex than simple text searching via Google when it comes to extracting answers from websites and databases. Fortanix is trying to build a security wall to protect search queries and data extraction from AI systems.
Generative AI technologies will eventually provide granular answers based on the associativity of information and retrieve information from sources that users may not be aware of, says Richard Searle, vice president of confidential computing at Fortanix.
“What we’re finding… is that in the AI space there’s more focus on privacy and information consent consent. Those are obviously the main cases,” Searle says.
Fortanix is effectively building a security wall around how AI search queries are handled. Fortanix is building security around prompts, where users provide AI search queries. The security wall extends to data recovery from LLMs, which are delivered to customers.
“We think there’s a significant market there. The partners we work with in the AI space are already talking about research to their customers today,” Searle says.
Fortanix’s AI private search is tied to confidential processing, which creates secure vaults accessible only to authorized parties via keys. The data is processed internally without leaving the vault.
“Data will be subject not only to the data protection regulations that we have today, but also to these emerging regulations around artificial intelligence,” Searle says.
Fortanix’s technology is a component in the emerging field of private artificial intelligence. Confidential AI is being extended to AI scenarios typically associated with GPUs and accelerators, said Mark Russinovich, chief technology officer of Microsoft Azure, during a roundtable at the Open Confidential Computing Conference in March.
Vector database protection
Private search with AI relies on data extraction from knowledge graphs or vector databases, which could draw information from a wide range of sources, including static conventional databases.
At Nvidia’s GPU Technology Conference, company CEO Jensen Huang defined vector databases as a new style of database that accepts structured or unstructured data that can be reindexed by encoding the meaning of the data.
“Now this becomes an AI database, and in the future, once it’s created, you can talk to it,” Huang said.
Fortanix’s goal is to ensure privacy for the search initiator – a human or machine user – and to protect the privacy and integrity of information that may be within the embedded vectors.
“I think confidential computing has a very important role to play,” says Fortanix’s Searle.
Private search is different from conventional search
Confidential AI prevents AI information from leaking, helping you meet regulatory requirements. The layer also anonymizes information so that the user’s intent or identity is protected. This is the opposite of conventional search, where user information and motivations drive Google’s ads and analytics.
A higher level of AI privacy is a cornerstone for industries such as healthcare and banking, which are tied to regulatory issues.
“Imagine you want to carry out research in a scientific domain: you may want to be able to use data from an institution other than the one you work in that has specific expertise in a particular field. Showing the privacy of that data and l The accuracy of this research is important,” Searle says.
Reserved AI status
Private artificial intelligence is an emerging concept and the research fits this profile. Intel and AMD have chips with hooks for secure enclaves. Microsoft, Google, and Amazon provide reserved computing virtual machines in their cloud services.
Companies can bring together datasets to effectively train the fundamental model, and these datasets are becoming high-value assets for leading companies, said Ian Buck, vice president and general manager of Nvidia’s hyperscale and high-performance computing division , during the round table.
The IT industry is rapidly moving towards deploying confidential computing in data centers, edge computing and PCs, but there is much more to do with confidential AI, said Greg Lavender, Chief Technology Officer at Intel, during the panel.
“Privacy, security and responsible AI will become important functions in this context, as the government adopts AI technologies from industry,” Lavender said.