Price:

Rob Hamilton warns Chinese AI models threaten Bitcoin code

Aug 14, 2026Summary from 4 podcasts.
  • Coldcard firmware flaws enabled automated scripts to drain over 1,400 Bitcoin.
  • Open-weight AI models like Kimi K3 now scan public repositories for zero-day exploits.
  • Marathon's private transaction relay rescued 5,600 Bitcoin from public front-running bots.

In August 2026, cheap AI models exposed open-source software's darkest secrets.

A five-year-old firmware flaw in Coldcard devices turned key generation into a guessing game. On Ungovernable Misfits, researcher Orange Surf explained that a 2021 compiler bug bypassed hardware randomness, forcing wallets to generate keys using predictable software entropy. Without manual dice rolls, generated seed phrases were trivial to brute-force. Attackers swept over 1,400 Bitcoin from exposed wallets almost overnight.

The crisis rapidly broadened as bad actors used uncensored AI models to audit public codebases. On Stacker News Live, hosts Keon and Carr detailed how automated scripts probed critical infrastructure like BTCPay Server, forcing swap provider Bolts.exchange to permanently shut down under a barrage of exploits. AnchorWatch CEO Rob Hamilton responded by forming the Bitcoin Red Team, spending tens of thousands of dollars running Chinese model Kimi K3 against 500 open-source repositories to patch flaws before hackers exploited them.

On The Jack Mallers Show, Jack Mallers argued that public source code without commercial backing creates a dangerous security illusion. Coldcard's restrictive licensing prevented outside firms from building products on its software, leaving zero financial incentive for third-party audits. Contrast that with permissive projects like Bitcoin Core, where massive institutional capital funds continuous security battle-testing.

"Without an active review economy, public source code simply hands attackers an easy map."

- Jack Mallers, The Jack Mallers Show

Mallers also criticized American AI regulations for hobbling white-hat defenders. Closed-source US labs like Anthropic build regulatory moats and enforce usage blocks on security research, forcing domestic teams to rely on Chinese open-source models. Black-hat attackers ignore policy guardrails entirely while patriotic researchers face artificial friction.

On-chain emergency operations quickly followed. Orange Surf tracked over 5,600 Bitcoin rescued through Marathon's Slipstream private transaction service, which allowed 2-of-3 multisig holders to migrate funds without exposing uncompromised keys to public front-running bots.

"Network side-channels turned out to be an essential safety net when public mempools became predator territory."

- Orange Surf, Ungovernable Misfits

A week of mounting evidence shattered the informal trust model surrounding hardware vendors. On Ungovernable Misfits, co-host Q noted intense fallout over Coinkite leadership deleting old posts and dismissing early warnings, alongside revelations that lead developers approved their own code using secondary pseudonyms.

Manual audits are officially dead. Open-source transparency only works when defenders automate attacks faster than their adversaries.