Erin Redwing warns AI exploits shatter Coldcard security
- AI scanners exposed entropy flaws in Coldcard wallets, draining 1,778 Bitcoin.
- Stolen funds moved to fake recovery sites as scammers target victims.
- Red teams now use AI models to audit open-source code defensively.
The self-custody myth just broke.
An unaligned AI model called Kimi K3 exposed a fundamental entropy flaw in Coldcard hardware wallets that human auditors missed for years. On TFTC on September 30, 2026, guest Erin Redwing described the breach as a paradigm shift for Bitcoin security culture. Automated AI tools are now scanning open-source financial code around the clock, uncovering deep cryptographic vulnerabilities across hardware wallets and Lightning Network infrastructure before human maintainers can patch them.
The exploit compromised over 8,600 addresses and drained 1,778 Bitcoin from air-gapped devices once considered impenetrable. Three days earlier, on September 27, 2026, Stacker News Live host Carp reported that 52 stolen Bitcoin moved into a single address linked to a public recovery trust domain. Co-host Keon and security researcher Kevin Loaec warned that the domain is almost certainly a secondary phishing scam designed to trick desperate victims into revealing private keys or personal details.
For years, Bitcoiners believed open-source, Bitcoin-only hardware was inherently safe. That hubris is gone.
Host Marty Bent noted on TFTC that the attack forced developers into a sobering retrospective, leaving even experienced users hesitant to recommend standard self-custody setups. In response, independent red teams are deploying the same automated AI scanners to audit core repositories and harden software defensively. Open-source monetary protocols sit on the front lines of automated exploit scanning, acting as a canary in the coal mine for the broader financial system.
The incident has also exposed a sharp rift over AI safety philosophy. On TFTC, Redwing argued that corporate leaders in Silicon Valley exploit existential dread and doomsday rhetoric to push for central licensing regimes and regulatory moats. While corporate executives demand administrative control over computing infrastructure, Redwing insisted that practical containment engineering and open-source defensive auditing offer far better protection against automated attacks.
Self-custody still offers sovereignty, but absolute hardware safety is dead.