Long videos.
Clear notes.

Main points

• Some older Coldcard Mark III wallets reportedly generated private keys with far less randomness than intended—around 32 bits of entropy rather than a 256-bit space. That could make affected keys searchable using ordinary computing hardware.
• The speakers associate the issue with the movement of roughly 1,000 BTC, but the connection between the suspected vulnerability and every transfer has not been definitively established.
• An open-weight model called Kimi K3 reportedly produced a detailed vulnerability report, while guarded closed models initially restricted similar analysis.
• AI can reduce an initial security audit from weeks and large specialist budgets to hours and hundreds of dollars. Its findings still need expert validation and reproducible testing.
• Domain expertise combined with AI helps defenders reject false positives and understand which findings are genuinely exploitable.
• Some companies are building continuous, locally hosted “AI immune systems” that review code changes, dependencies, and repositories without sending sensitive findings outside the organization.
• The incident does not invalidate self-custody. Multi-vendor multisignature setups can reduce single-device risk, although they add operational complexity.
• There is no universal custody answer: self-custody, professional custody, and Bitcoin ETFs trade independence against convenience, privacy, liquidity, and counterparty risk.
• AI may weaken the market for undisclosed zero-day vulnerabilities because flaws can be found and exploited faster, forcing maintainers toward continuous security review.
• The quantum-computing section is speculative. The speakers believe useful cryptographic attacks remain years away but argue that Bitcoin should prepare migration paths.