Bitcoin Forum

Is anyone aware of a project that's working on this?

can you be more specific? ???

He's probably referring to autonomous agents.

Yeah, sorry autonomous agents.

The wiki refers to them as plain "agents"

The Achilles heel of autonomous agents is their inability to hide the private key from parasitic hosts.  The Wiki talks about using TPM as a solution, but that is a very weak solution because TPMs can be reverse engineered (dissected in an electron microscope if need be) and simulated in a VM.  Even if the amounts of money are small, hackers are going to attempt this as a matter of honor.

Agents are only going to be truly viable once homomorphic encryption has matured.  But this is still a nascent technology, so it's going to take several years before agents take off.

A fully fledged Bitcoin Agent is very far ahead in the future. I am working on a rudimentary version of StorJ.

I think it's easy to over-generalise about the godly power of "hackers" in the abstract. Individual TPM chips may or may not be weak, although fwiw the future of TC is probably entirely on-die systems like Intel SGX. As far as I know no SEM-wielding hacker has ever reverse engineered modern Intel chips, the difficulty of that is of the level that it'd likely take an entire team of highly trained and highly paid people, even then they may not get anywhere.

TC is not likely to be the weak point for agent development in the forseeable future. There are many ways to scam/kill agents that don't rely on hacking their hosting environment.

I referred to self-replicating autonomous artificial intelligent agents as described in:
https://en.bitcoin.it/wiki/Agents

--

I enjoyed your talk at
https://www.youtube.com/watch?v=Pu4PAMFPo5Y, sir.

:)

Wikipedia : An autonomous agent is an intelligent agent.
WTF is an agent ?

The embedded stuff quite difficult. I mean just look how hard it is build these ASICs. Indeed there will be many ways to scam/kill agents, but the network must/should be designed to deal with that.

I wonder if you could use some sort of Proof-of-Storage system instead. Obviously you want these agents to be as portable as possible, and not need specialized hardware to run.

Because there is nothing interesting to discover there. x86 is pretty much an open standard.

In contrast, how long did it take until "unbreakable" Blu-Ray encryption was broken?



...or a swarm of thousands of unpaid undergraduate students cranking the ratchet. 

On the contrary, modern Intel chips contain signing keys for microcode, random number generators (but are they random?), on board TPMs (in the northbridge) and so on. There's quite a lot to investigate there. It's just very hard.

I'm quite familiar with Blu-Ray content protection, by the way. Blu-Ray was not designed to be unbreakable, fevered fantasies of some MPAA execs aside. It was designed to be broken and resealed continuously throughout its expected lifespan. In the end the HDCP key leak put an end to that plan (still nobody seems to know where it came from), and more generally the problematic dynamic that so often kills DRM schemes where the people who make players have backwards financial incentives, resulting in a box-ticking approach.

A better comparison might be the Xbox 360 which has been under highly skilled and sustained assault for nearly a decade. It's been hacked and resealed a few times over time. The back and forth was very interesting to watch. I don't remember what the current state is though.

Anyway, it's really no understatement to say that very few people have both the skills and the interest to  engage in hardware hacking, and they are rarely motivated only by money. If you follow these areas as closely as I do, you'll see the same names crop up over and over again.

On the other hand, I'm not aware of any "secure" chips that have stood up to inspection by FIB (http://en.wikipedia.org/wiki/Focused_ion_beam).  Search for Tarnovsky on youtube for details.  All "security features" so far are annoyances and speedbumps rather than roadblocks.

I'm not sure if a move to the main die will help or not.  I think that amateurs are mostly working on small run, old fab, large feature chips.  Does anyone know for sure that Intel-sized features are fundamentally beyond the reach of a FIB?

Totally agreed that the real threat will be on the software side though.  For one thing, good system design should be able to prevent "Break Once, Run Everywhere" attacks, so the theoretical capability to extract keys from a TPM isn't a big deal, particularly not for distributed/networked agents.