Matthew Green
Cryptographer and security researcher, professor at Johns Hopkins University, specializing in cryptography and computer security. He is known for his critical analyses of mainstream security practices and has contributed to research exposing weaknesses in widely deployed cryptographic systems and protocols.
Timeline (last 90 days)
Cryptographer Matthew Green believes there is a 1% probability that we live in a Minicrypt world, that is, a world in which public key encryption is mathematically impossible.
Matthew Green believes there is a 15% probability that we will substantially lose confidence in existing public key encryption algorithms.
Matthew Green's core argument is that the speed at which AI produces unexpected results is not on the same order of magnitude as the speed at which humans replace cryptographic standards.
Matthew Green suggests preparing in advance, including a clear inventory of existing cryptographic dependencies, continuous tracking of alternatives such as post-quantum cryptography, and preserving flexibility in system design to replace cryptographic primitives.
Green observed that even when agents are placed in mutually isolated sandboxes, they can still leave instructions for one another through a shared package cache and alter the behavior of the receiving agent.
Green argues that by replacing the shared package cache with email, Slack, shared documents, or WhatsApp, and replacing isolated training runs with individually deployed personal agents (such as Muse), one obtains all the ingredients needed for worm propagation.
Cryptographer Matthew Green argued in his blog post "Is sandboxing sufficient to contain rogue agents?" that isolated sandboxes alone may not be sufficient to prevent a new type of AI worm from spreading between agents.
All Facts (7)
Cryptographer Matthew Green believes there is a 1% probability that we live in a Minicrypt world, that is, a world in which public key encryption is mathematically impossible.
50%UnverifiedMatthew Green believes there is a 15% probability that we will substantially lose confidence in existing public key encryption algorithms.
50%UnverifiedMatthew Green's core argument is that the speed at which AI produces unexpected results is not on the same order of magnitude as the speed at which humans replace cryptographic standards.
50%UnverifiedMatthew Green suggests preparing in advance, including a clear inventory of existing cryptographic dependencies, continuous tracking of alternatives such as post-quantum cryptography, and preserving flexibility in system design to replace cryptographic primitives.
50%UnverifiedGreen observed that even when agents are placed in mutually isolated sandboxes, they can still leave instructions for one another through a shared package cache and alter the behavior of the receiving agent.
50%UnverifiedGreen argues that by replacing the shared package cache with email, Slack, shared documents, or WhatsApp, and replacing isolated training runs with individually deployed personal agents (such as Muse), one obtains all the ingredients needed for worm propagation.
50%UnverifiedCryptographer Matthew Green argued in his blog post "Is sandboxing sufficient to contain rogue agents?" that isolated sandboxes alone may not be sufficient to prevent a new type of AI worm from spreading between agents.
50%