Anthony Brooks

@Johnkurrency @m477lander I don't disagree tbh, separation helps, but only if you realize a lot of this systems still fail from the same trust boundary. That is where things usually fall apart.

Anthony Brooks

@Anthonybrooks_b

3 days ago

Up and grateful to the Lord 🙏

Anthony Brooks

@Anthonybrooks_b

3 days ago

@m477lander imo the boundary should sit at the point where external data becomes verified by local state.

193

Anthony Brooks

@Anthonybrooks_b

3 days ago

Actually, imo boundary should sit at the point where external data becomes verified local states.

Matt lander

@m477lander

4 days ago

Yeah, this is pretty much the whole problem. The question is not really where to keep BTC anymore. Every option just moves trust and risk into another layer. A CEX can freeze withdrawals or blow up. A hot wallet is only as safe as the device it runs on. A hardware wallet depends on firmware, keygen and the manufacturer not screwing up somewhere deep in the stack. And the second you actually use the asset, you inherit bridges, oracles, RPCs, contracts, external signers and the whole execution path. You can solve storage and still lose everything on the first transaction. I have been off the timeline for a month and a half. The industry apparently used that time to keep bleeding money. 30 major incidents. More than $210M gone in July alone. The number is not even the interesting part. In the clearest cases, nobody hacked the blockchain. The chain did exactly what it was told to do. The security model around it failed. Coldcard, @COLDCARDwallet. A Bitcoin only hardware wallet people trusted for years because it was supposed to be the safest layer. Affected firmware could fall back to deterministic software seed generation instead of proper hardware RNG. Effective entropy could drop to around 40 bits on Mk2 and Mk3, and around 72 on newer affected models instead of 128. After that, the attack is almost mechanical. Brute force candidate seeds offline, derive addresses, check them against the chain, drain the funded ones. No physical access needed. The first major wave took 1,082.65 BTC from 1,196 addresses in 41 minutes. Around $70M at the time. Cold storage that was never actually cold. No stolen device. No phishing. No user signing random garbage. The seed was weak from day one. That is the Keygen TCB. The lowest layer. The exact layer I have been talking about for a long time. Once that breaks, everything above it is theater. AFX, @AFX_XYZ. Seven validators. 2/3 threshold. Multisig. Dispute window. Every box checked. The attacker got signing authority over five validators, enough for 7,142 out of 10,000 voting power and a fully valid withdrawal. The quorum was real. The signatures were valid. The contract was untouched. After the 200 second dispute window, the bridge released $24.15M USDC exactly as designed. The multisig did not fail. The trust model did. The supposedly independent keys were sitting inside the same operational risk zone. Ostium, @Ostium. Same problem, one layer higher. The attacker got access to an authorized oracle signer and a registered PriceUpKeep forwarder, then pushed correctly signed reports with fake prices and controlled timestamps, including future dated data. Those reports were used to open synthetic BTC/USD positions against the vault. The verifier checked whether the signer was authorized. It did not check whether the data made sense. The OLP vault lost $23.75 M. Cryptography proved who signed the data. It did not prove the data was true. Different products. Different attack surfaces. Same architectural mistake. Control of a key is not trust. A valid signature is not a valid action. An authorized source does not make the data true. Blind signing is the user facing version of the same problem. A wallet can prove that a key signed a set of bytes. It cannot prove that the economic result matches what the user intended. And the industry keeps responding the same way. One more audit. One more multisig. One more scanner. One more policy layer. One more check bolted onto the same old architecture. But when keygen, risk verification, signing, routing and execution live in separate trust zones, every boundary becomes a place where context gets lost and extra trust slips back in. So the answer to the original post is not another hardware wallet. And it is definitely not another wallet with a swap button. This market needs a different category. An isolated execution environment that stays fully noncustodial. One hardened perimeter under user control, where keygen, local risk checks, typed intent, policy, signing and protected execution are tied together instead of scattered across separate trust zones. Everything coming from outside stays untrusted until it is checked locally. Calldata. RPC responses. Routes. Spender permissions. Token metadata. Protocol state. All of it. The user should not be signing opaque bytes. The user should be signing a clear intent bound to an active policy. The backend holds no keys, signs nothing and gets no delegated authority. I am tired of writing the same postmortem after every drain and explaining which key or trust boundary failed this time. Protocol names change. Numbers change. The exploit changes. The architecture underneath does not. At some point, repeating the same warning for the hundredth time stopped being useful. Building the answer made more sense. So I was not quiet for a month and a half because I had nothing to say. I may have just been busy building the answer to the question in this post.

228

241

55K