Opinion
Taking the hackers to a new battlefield they don’t know
"A faster algorithm is no help," writes CyberRidge CEO, Prof. Dan Sadot, "if the transmission never becomes a usable digital recording."
For an optical communications engineer, noise is usually something to overcome. The job is to send a signal through a fiber and recover its information at the other end. In my research, noise became part of the protection: the intended receiver must recover the information, while anyone listening along the route must be unable to preserve a usable copy.
Normally, an adversary can use a small optical coupler to copy encrypted traffic from a live fiber link without interrupting the connection. No code has to be broken to collect it. This is the harvesting gap: strengthening encryption makes the contents harder to read, but does not prevent the recording. Both ends of the link may see normal service. Successful delivery is not proof that no one else has collected the transmission.
Data centers serving AI workloads are being built in remote locations where land and power are cheaper. Training data and model weights moving to and between those sites cross fiber routes beyond the secured buildings. Junction boxes, street cabinets, and stretches of cable along isolated transport corridors offer physical access points that expand the attack surface. Choosing an economical location for computing infrastructure also makes the route to that infrastructure part of the security decision.
A recording makes harvest-now-decrypt-later attacks possible by letting an adversary separate the moment of interception from the work of decryption. The attacker can wait for new cryptanalytic methods, a sufficiently capable quantum computer, or AI agents that hunt for the keys. The organization may upgrade its encryption in the meantime, but it cannot upgrade the copy already in someone else's possession.
Genetic information can remain sensitive for generations, long after the systems carrying it have been replaced. Unlike a password, it cannot simply be changed after exposure. For an organization transmitting such information, the security horizon is the life of the secret. Decisions about what an interceptor can retain therefore have consequences that reach far beyond the network's next upgrade.
Security design determines what kind of problem an interceptor gets to solve. On a fiber link, preventing a usable recording changes the problem from digital processing that can be deferred to optical reconstruction that must happen as the light passes.
One way to do that is to manipulate the light using a constantly changing optical key every fraction of a second and to bury the signal in optical noise, well below the signal-to-noise ratio threshold theoretically required to reconstruct useful information. To a listener without that optical key, the transmission looks like background noise. Only the intended receiver, maintaining the optical key in real time, can reverse the optical manipulations and recover the optical signal above the noise before converting it into digital data.
Recording the intercepted light without first reconstructing it optically destroys the information at once and forever. The listener has not stored a tougher encrypted file; it has failed to capture the raw material needed to obtain one. More powerful digital processing cannot restore information that does not exist.
Access to the fiber therefore no longer completes the collection stage. The interceptor needs optical attack tools and the correct optical key at the moment the signal passes; acquiring either later does not turn the recording into usable material.
Post-quantum cryptography, by contrast, is designed to resist quantum attacks on public-key systems. It strengthens the mathematical protection of data; optical protection addresses the ability to collect a usable copy. Both layers can protect the same transmission. Keeping an interceptor from obtaining a recording complements the effort to make encrypted content harder to crack.
Communications engineers still have to deliver the information reliably. Asking an operator to accept a signal that looks like noise means challenging the instinct to keep that signal clean. It also means proving that the intended receiver can recover it through the network's amplifiers and other equipment. In the optical network laboratory of a leading European telecom operator, our system maintained a protected link across a 300-kilometer topology using existing network equipment.
A usable recording lets the adversary choose when to attack and which tools to bring. Denying that recording forces the interceptor to reconstruct the optical signal while the light is passing. Computing power acquired years later cannot make up for missing that window.
Prof. Dan Sadot is the founder and CEO of CyberRidge.
