Wilderminds

I think you are missing the point of what $ICP is actually solving. Let's look at some major hacks: Bybit: Had exactly what you prescribe: an air-gapped multisig cold wallet with hardware signers. Still lost $1.5B. The compromised laptop wasn't even Bybit's. A Safe Wallet dev was hacked, attackers obtained AWS credentials and replaced the JavaScript in Safe AWS S3 bucket. The signers approved what the screen showed them. The screen lied. The malicious JavaScript was removed two minutes later. Binance: DevOps laptop -> service container -> the signing system signed whatever it was told. No private key was leaked. Bangladesh Bank: Custom malware inside SWIFT Alliance Access, manipulating the Oracle DB to conceal fraudulent transfers. Capital One: A misconfigured WAF on AWS. See the pattern? The pattern isn't "humans". It's one human reaching one mutable server that sits between the signer and the chain. #InternetComputer removes that server. An attacker compromises a mutable component of the infra and changes what the system does. That infra is the attack surface, not the person signing. That's the attack surface $ICP is designed to eliminate: ✅Frontends served directly from canisters, with responses certified by the subnet and verifiable by the client. No S3 bucket where attackers can silently replace JavaScript. ✅No traditional service containers, SSH access or admin controlled databases. Canister state changes are executed through subnet consensus. ✅No single administrator holding the upgrade keys on a laptop. Canisters can be governed by a DAO, with upgrades requiring a vote. ✅No complete signing key sitting in one place. Threshold crypto distributes key shares across nodes, while properly implemented consent messages let users verify what they're signing. ✅ No API keys or cloud credentials. Canister calls are authenticated by the protocol, nothing for a compromised laptop to hand over. ✅ Verifiable deployed code. Wasm hash is onchain, reproducible builds let anyone check it matches source. ✅ Public controllers. Anyone can see who can upgrade a canister, one key, DAO, or nobody. ✅ Withdrawal policy enforced by consensus. Limits and approvals live in the canister, not in a backend someone can reach. ✅ Upgrades go through a visible voting window. A malicious change is a public proposal, not a silent push. ✅ Certified data reads. Balances and transaction lists are verifiable as coming from the subnet, not a tampered node. ✅ No runtime supply chain. Dependencies compile into the WASM, nothing fetched at execution time. ✅ No passwords or seed phrases to phish. Internet Identity uses passkeys, with a separate principal per app. ✅ No key server. vetKeys derive encryption keys onchain via threshold crypto. Your three "someone still has to" points: code bugs, key custody and user signing and those are real risks, and they apply everywhere. ICP has never claimed otherwise. What $ICP solves is that the infra layer exploited in hacks like Bybit, Binance, Bangladesh Bank etc.. can be fundamentally redesigned to eliminate entire classes of attack. #InternetComputer isn't trying to make traditional infra more secure. It's eliminating entire attack surfaces by making that infra unnecessary. That's why I'm genuinely optimistic about ICP future. The tech addresses real problems, and its potential extends far beyond what we've seen in crypto. I like the direction @dominic_w is taking on X lately, focusing on practical problems and what $ICP can do to solve them.