2 ms·
There are a variety of names. DAC is something Ethereum came up with. Before then they were being called autonomous agents: https://en.bitcoin.it/wiki/Agents h
by mike_hearn 12y ago
There are a variety of names. DAC is something Ethereum came up with. Before then they were being called autonomous agents:
https://en.bitcoin.it/wiki/Agents https://en.bitcoin.it/wiki/Agents
I was never totally clear on the difference, but as people often refer to Bitcoin itself as a DAC, I feel the term is too vague to be useful.
The next step to make these REALLY useful for agents is to set up Intel TXT on them. This would likely involve some voodoo as TXT is appallingly badly documented. But, if you make sure the hardware is compatible and that hardware virtualisation integrates properly using tboot it should be possible.
This step is a game changer because then an agent can get a hardware-rooted proof that the remote system is secure including against Kordless or whoever else owns the hardware. The agent then knows it's safe to install a copy of itself on the new machine, have the copy create a wallet, and send money to the new box such that the operator can't steal the money but the new agent can still pay for its own hosting costs.
Once done the agent would effectively be autonomous if it had a way to earn income e.g. reselling disk space. Scary stuff!
- al2o3cr 12y agoI didn't see anything in the TXT whitepaper about securing programs run by a hypervisor against attacks by the hypervisor. Seems more focused on making sure that managed VMs can't compromise their neighbors or do side-channely things. If I provide your VM with a simulated "TPM" that actually emails me the private keys, how is that supposed to be secure?
- liam8841 12y agoIf I understand rightly, this sort of security is impossible until fully homomorphic encryption is invented. Any actual cryptos available to comment?
- mike_hearn 12y agoThe TPM is un-simulatable, that's the purpose of it. The hypervisor is included in the remote attestation. The theory is, if you trust the signing keys used in the TPM (i.e. Intel/the hardware), then you can get a message from the remote computer that attests what code is running including the hypervisor. Thus you can conclude that the remote environment is safe against all but quite sophisticated hardware attackers (assuming a modern setup where you can lock cache lines and the TPM is integrated with the northbridge).