A one-way function can make recovering a secret computationally infeasible, but it doesn’t eliminate the implications of allowing the device to decrypt its own data.
I presume you’re referring to a workflow where a password is put into the device, fed through a one way function to derive the key, then the key is used. So the key is derived and never stored on disk.
Nonetheless, your key was in memory. There’s an attack surface for any forensic team. There’s also the capability of trying to brute force the password in a virtual environment after having cloned the device. FBI did this to an iPhone that had full reset after 10 failed attempts.
There’s also the capability to compromise the OS, leave the device there, wait for an admin to come by and supply the password themselves.
I don’t really see how the manner by which you derive your key affects the fact that having the key on device is a vulnerability. It may mitigate the vulnerability but you’re talking about corporations and government entities with practically unlimited budgeting. If the vulnerability exists, it will be explored.
I still think my points stands. It would be stupid for someone committing espionage to allow the same device to also decrypt the data.
You originally asked what the point is of a self encrypting device. Even assuming this particular device in question has the attack vector you pointed out (I don’t know, I haven’t checked), it isn’t pointless.
Your attack vector assumes the key is in memory. That would be a very odd and weird design that I’ve never seen… The auth key is usually stored on cryptographically secure hardware (a tpm), you unlock it at boot and it talks to the drive, the OS doesn’t know about it or see it.
There’s technically an attack vector there, an unproven one that assumes they have a backdoor or an undisclosed flaw at the hardware level… so sure, if you really really don’t want to risk it, encrypt your data yourself.
There are a lot of reasons to choose these devices:
Performance
Nothing to configure os side, you just write to it and it’s encrypted. Power loss means it cannot be decrypted by a malicious actor.
Disposal. You can often issue a single command that scrambles the auth key and now all the data is unrecoverable without you needing to write every byte.
Yeah totally fair, thanks for the corrections. I’m not really all that exposed to this field, I was feeling a little frank about the details but I was trying to make a larger point.
TPM can make it infeasible to extract the key, but there’s also the attack surface of bugging the bug. Once access is authorized to encrypted storage, protection depends on what happens outside TPM. That’s one. There’s also the potential to sniff the password, which would be two. They have full access to the device and room, with capacity to wait for the culprit to return for fetching data.
Correct me if I’m wrong but I also think drive encryption with TPM would require that the drive be decrypted while in use. This device wouldn’t be of any value if it weren’t in use, so there’s also that.
It seems like something you can probably count on being secure for something like a home security system. If running an espionage campaign against a trillion dollar company with Uncle Sam in the back pocket, I’d confidently bet against this system.
Not trying to make grandiose claims about the security of TPM. I just think this scenario was a particularly risky adventure — one which could have feasibility been easier to perform with less risk, had the device come with no capability to decrypt its own data.
I’m imagining something that perhaps micro-batches audio through a raw_audio - encrypt -> storage process, but say every 15 seconds or so. Then transport the encrypted data to a device with decryption keys.
Honestly I wouldn’t even waste my time. That guy seems like a complete asshole, who wants to sound smart but doesn’t even understand what you’re saying.
Encryption being a one way algorithm means nothing in the context of your original concern.
Sigh
https://en.wikipedia.org/wiki/One-way_function?wprov=sfla1
A one-way function can make recovering a secret computationally infeasible, but it doesn’t eliminate the implications of allowing the device to decrypt its own data.
I presume you’re referring to a workflow where a password is put into the device, fed through a one way function to derive the key, then the key is used. So the key is derived and never stored on disk.
Nonetheless, your key was in memory. There’s an attack surface for any forensic team. There’s also the capability of trying to brute force the password in a virtual environment after having cloned the device. FBI did this to an iPhone that had full reset after 10 failed attempts.
There’s also the capability to compromise the OS, leave the device there, wait for an admin to come by and supply the password themselves.
I don’t really see how the manner by which you derive your key affects the fact that having the key on device is a vulnerability. It may mitigate the vulnerability but you’re talking about corporations and government entities with practically unlimited budgeting. If the vulnerability exists, it will be explored.
I still think my points stands. It would be stupid for someone committing espionage to allow the same device to also decrypt the data.
You originally asked what the point is of a self encrypting device. Even assuming this particular device in question has the attack vector you pointed out (I don’t know, I haven’t checked), it isn’t pointless.
Your attack vector assumes the key is in memory. That would be a very odd and weird design that I’ve never seen… The auth key is usually stored on cryptographically secure hardware (a tpm), you unlock it at boot and it talks to the drive, the OS doesn’t know about it or see it.
There’s technically an attack vector there, an unproven one that assumes they have a backdoor or an undisclosed flaw at the hardware level… so sure, if you really really don’t want to risk it, encrypt your data yourself.
There are a lot of reasons to choose these devices:
Yeah totally fair, thanks for the corrections. I’m not really all that exposed to this field, I was feeling a little frank about the details but I was trying to make a larger point.
TPM can make it infeasible to extract the key, but there’s also the attack surface of bugging the bug. Once access is authorized to encrypted storage, protection depends on what happens outside TPM. That’s one. There’s also the potential to sniff the password, which would be two. They have full access to the device and room, with capacity to wait for the culprit to return for fetching data.
Correct me if I’m wrong but I also think drive encryption with TPM would require that the drive be decrypted while in use. This device wouldn’t be of any value if it weren’t in use, so there’s also that.
It seems like something you can probably count on being secure for something like a home security system. If running an espionage campaign against a trillion dollar company with Uncle Sam in the back pocket, I’d confidently bet against this system.
Not trying to make grandiose claims about the security of TPM. I just think this scenario was a particularly risky adventure — one which could have feasibility been easier to perform with less risk, had the device come with no capability to decrypt its own data.
I’m imagining something that perhaps micro-batches audio through a
raw_audio - encrypt -> storageprocess, but say every 15 seconds or so. Then transport the encrypted data to a device with decryption keys.Honestly I wouldn’t even waste my time. That guy seems like a complete asshole, who wants to sound smart but doesn’t even understand what you’re saying.
Encryption being a one way algorithm means nothing in the context of your original concern.