In brief
- Coinbase, Block, BitGo and 40-plus firms put their names to an open letter demanding frontier labs open trusted-access programs to qualified defenders.
- The Bitcoin Policy Institute organized it; signatories include Blockstream, ARK Invest among others.
- The gap is practical: the people patching wallets and node software can't reach the best models, while the people attacking them already can.
More than 40 bitcoin and crypto firms asked the biggest AI labs this week to let independent security researchers use their strongest models before public release.
The letter, published August 10, rests on a simple asymmetry. The Bitcoin network secures more than a trillion dollars, and the wider stack—wallets, signing devices, libraries, custody systems—holds trillions more. That code is written in the open, and one bug can erase someone's savings.
Yet the maintainers keeping it running can't get into the vetted researcher programs the AI labs run. When they reach for a public model, the guardrails shut down a legitimate hunt for flaws. So they lean on open-weight systems they can run themselves, which is useful, but a step behind the frontier—and the hackers attacking cryptocurrency networks.
“It is clear that early access and reasonable compute budgets would help core developers produce more secure code and identify and remediate vulnerabilities before losses occur,” the letter reads. “We, therefore, ask frontier AI labs to establish or expand standing trusted-access programs for qualified defenders of open-source financial infrastructure.”
The letter points out labs and a few partners get a months-long look at new attack capabilities before the rest of the world, and those capabilities escape anyway through public releases and compromised systems.
The labs haven't committed #
Late last month the Bitcoin Security Consortium—BlackRock, Coinbase, Strategy, Anchorage, ARK, Block, Blockstream, Fidelity Digital Assets and Galaxy—committed $15 million to the network's long-term security. The letter targets a different gap: not money for tomorrow's threats, but today's models for the maintainers patching them.
As of today, the big AI labs have not shared any public stance on this specific topic, however, the companies are working on better safeguards that would prevent their models’ abilities to be used for malicious intentions, no matter if it’s for blockchain coding or any other usage.
Anthropic has shown how the same models power "vibe hacking"—criminals running live Bitcoin-ransom attacks with AI help—and says the tools make offense cheaper. Giving defenders stronger models means picking who qualifies and trusting them not to leak.
The letter asks the labs to take that risk anyway.