AI has generated a level of distrust and dislike for any technology that would have been unimaginable 5 years ago. Job losses and total insensitivity to all other community issues generated by AI have made it far worse.
AI security, in particular, is now a true bugbear and making way too many headlines. Anthropic’s suddenly-famous Mythos Fable 5 is all about current and emerging AI risks.
The other glaring problem is less obvious. The AI sector, in its total lack of wisdom, is doing absolutely nothing to reassure civilization as a whole about AI. The problems are obvious everywhere and the fixes are invisible.
This is where confidential computing rather modestly rides in, not on a white stallion but on the odd NVIDIA press release about NVIDIA partnering with Apple’s Private Cloud Compute.
It’s like finding the Cure for All Stupidity in the weather report.
You really do have to wonder who’s looking out the window when the AI storms blow in.
Confidential computing basics
As a rule, you never hear about what goes right on any subject in the news. Whatever’s going right about AI, if anything, let alone AI security, seems to be a non-topic.
Confidential computing has, in fact, been ticking away in the background. It hasn’t exactly been roaring out of the headlines.
The big perceived threat from AI is a super-bot able to breeze past conventional security. This relates to everything from basic privacy to financial management. That’s more or less the wrong planet for the actual threats. It’s the core data that’s at risk.
Confidential computing just happens to be based on protecting data when it’s in operational. When it’s most vulnerable and most revealing in the form of transactions, for example, or when using key security data entry.
If you’re seeing a glaringly obvious fix for managing the most basic elements of security and data when it’s most at risk, bingo. Thankfully, NVIDIA does have some specific information about this process.
Confidential computing comes with some terminological baggage, but it’s pretty straightforward. “Trusted execution environments” are the security mode and processes that confidential computing addresses. Unauthorised parties may not access or modify code in use or after executing.
These safeguards are called “isolation” and may apply to virtual machines, applications and functions. It would be almost impossible to get around these in-process security measures.
These security measures can be operated on CPUs and GPUs. There doesn’t seem to be any real limit to the range of confidential computing measures you can put on a system.
AI security 101 and the no-brainer of all no-brainers
The idea of stopping executable intrusions by permissions isn’t new. It’s been around for at least a decade. Microsoft introduced it as an OS feature circa Windows 7.
Stopping the strikes while in active process is new. It’s a genuine threat to malware functions. It’s very practical. AI has raised the security bar to unheard-of values.
Confidential computing could well be the broad-spectrum fix for just about everything that’s wrong with global cybersecurity on so many levels. It can be used for everything from national security to basic shopping. The current state of damage from cybersecurity and cyberespionage is at an all-time high. Compromised security is now a global industry, and it’s all AI-driven.
Confidential computing could be the best possible selling point for managing security.
The PR image catastrophe that AI’s turning into can be shut down with two words, “confidential computing”.
People don’t want to have to worry about their super-valuable IP getting hijacked character by character while it’s in process. They don’t want their possible years of work on brand-new stuff getting cloned 5 seconds after they make it.
How hard can this possibly be to understand? This image problem is critical to the adoption of future AI in all its forms. A real fix must be highly visible. People talk about “consumer confidence” as if it were a given. It isn’t, and with the current chaotic state of AI introduction, it is absolutely essential.
When does Confidential Computing hit the mainstream?
Getting confidential computing into the mainstream is the crucial next move. It’s happening at the corporate and Cloud levels. Apple, NVIDIA, and Google have been solidifying their security technologies into a working framework.
This has probably been going on for some time, and the higher levels of interoperation on AI security understandably haven’t been visible. You don’t have to explain your security measures, but they have to be like a good guard dog, where you can see them.
The perceived dangers of AI and its many actual disasters have made it critical to ensure that AI security is as common and understandable as SSL for mainstream use.
At consumer level, confidential computing must be an everyday thing you can use and feel secure about using. It needs to reassure businesses that the security of their highly vulnerable systems are actually under real-time management. It looks like confidential computing is ready to roll.