Secure encryption is one of those background utilities we all depend on, like plumbing and sewers: it only comes to our attention when something is going wrong. David Clinton says that things are going wrong with encryption standards, thanks to the technical omnicause, artificial intelligence:
Without secure encryption, we’ll all be living dangerously
This topic has a guaranteed boredom score of 12 (where “10” represents catatonic). But its could make a difference to my own life score is 15 (where “10” represents “Yikes!”).
Data encryption is the peg on which the possibility of all modern financial and healthcare-related activity hangs. Without standard encryption algorithms like AES and RSA:
- The contents of every exchange of information between you and your bank’s website would be visible to anyone on the internet.
- The passwords you use to authenticate to your digital accounts would be readable by anyone with access to any node between you and your account’s host.
- The sensitive data stored by any financial, healthcare, or government organization would be visible to anyone with access to their host systems — access made possible by decrypted passwords (see above).
Encrypted data contains all the original information. But its value lies in the assumption that the underlying algorithm is so complex that it would require the use of powerful computers and hundreds of years to crack it.
For the past 50 years, that assumption has held up well. But the theoretical existence of quantum compute resources — which can execute exponentially more compute operations than traditional processors — presents a theoretical threat to existing algorithms.
In other words sometime within the next 10-20 years, quantum compute devices will be powerful enough to brute-force the end of modern encryption. That sounds scary.
But it’s arguably not even the real problem. That, as Tyler Cowan recently wrote in the Free Press, would be the recent explosion of AI-driven discoveries in mathematics. Such events have introduced concerns about the odds encryption security could be compromised in months rather than years.
Frontier AI models are now at least making progress solving longstanding mathematical problems. The pace of discovery seems to be growing and it seems inevitable that someone somewhere will turn a powerful AI model on the algorithms we rely on.
