40+ crypto firms want early AI model access for vetted researchers, citing a 4,962-finding audit and an exploited BTCPay flaw. The letter stays open for signers.
More than 40 bitcoin and digital-asset organizations signed an open letter on Aug. 10 asking leading AI labs to give vetted security researchers controlled early access to frontier systems. Attackers, the letter argues, are gaining an unfair advantage.
The coalition wants standing trusted-access programs for researchers who show legitimate security responsibilities, with early entry to advanced models and prerelease systems when appropriate. Frontier AI changed the economics of security work, and the letter lays out why. Advanced systems scan large collections of code for weaknesses and test whether a flaw is exploitable. Legitimate security review gets faster and more thorough. Attacks get cheaper and easier to scale.
Bitcoin Policy Institute published the letter, titled "Defenders Need the Frontier." The group spans exchanges, custody firms, wallet makers, asset managers and bitcoin miners. Signers include Coinbase, Bitgo, Block, Blockstream, Anchorage Digital, ARK Invest, Bitwise, Foundry, Strategy, Galaxy and Trezor. Developer-support groups Brink, Chaincode Labs, Btrust, OpenSats and BTCPay Server also signed. The miner MARA holds a 19/100 Alpha Score at AlphaScala, a Weak reading.
Large labs and a small circle of partners see those capabilities early, the coalition argues. People who maintain widely used open-source financial software wait for general release. By then, attackers may already have accessed or copied comparable capabilities, or built their own, the letter argues.
"Frontier AI is changing the economics of both security research and cyber operations," the letter says. Defenders routinely hit safety restrictions when they try to use public AI products for legitimate research, leaving them "to rely on less capable open-weight alternatives for critical security reviews." Open-weight models, whose underlying code can be downloaded and adapted, help small teams get work done. The coalition says they do not match the strongest systems the labs keep under controlled access, and the gap matters when a small team inspects software that moves real money.
Pressure is already visible among small maintenance teams, the letter says. Bitcoin Policy Institute has received reports that sophisticated actors, including possible foreign adversaries, are using advanced AI to keep pressure on open-source developers. Even unsuccessful attacks burn time and money that maintainers would otherwise spend improving their software.
Bitcoin alone secures more than $1 trillion in value, according to the letter. The broader digital-asset system runs on open-source wallets, payment processors, exchanges and Lightning Network tools. Cryptographic libraries and custody services run on the same open-source base. A flaw in any of those pieces can have serious consequences. Digital assets function like bearer instruments: control of the cryptographic credentials means control of the funds. Once money moves, recovery is often impossible. A security weakness can expose retirement accounts, business funds, emergency savings and institutional capital, the letter says.
Coldcard's recent breach fits the pattern. AI was suspected of spotting a flaw in the hardware wallet's firmware before the issue became public.
In early August, a volunteer project called the Bitcoin Red Team used AI-assisted tools to audit bitcoin-related code. The group covered 390 projects in about 27.5 hours and filed roughly 4,962 findings, dozens classified as critical and hundreds as high severity, according to an early snapshot posted to X. Participants included Cashu developer Calle and Anchorwatch CEO Rob Hamilton. The team worked across Moonshot's Kimi K3 and Anthropic's Claude variants, with OpenAI systems also in the mix. OpenSats helped fund compute costs reported in the tens of thousands of dollars. Access limits forced the group to lean on whatever models were available for the initial pass, the letter says.
None of those findings are confirmed vulnerabilities. Each report still requires an experienced person to verify it and assess severity. Coordinating the fix with a project's maintainers comes after. Close to 5,000 findings in under two days showed how quickly AI can help researchers sort through complex code.
The same flood shows up at Coinbase. Bug bounty volume is on track to triple, and AI-generated submissions now fill the review queue.
BTCPay Server, open-source software that helps merchants accept bitcoin payments, disclosed a critical flaw affecting versions before 2.4.2. The vulnerability let an unauthenticated remote attacker obtain sensitive administrator credentials for LND, a commonly used Lightning Network implementation. Those credentials would give an attacker control of connected wallets and allow funds to be drained. The flaw was actively exploited, and some operators reported losses.
Sparrow Wallet developer Craig Raw played a key role in identifying the issue after being affected, and Bitcoin Red Team members helped analyze it. BTCPay said AI is changing the balance between attackers and defenders by lowering the cost of reviewing large codebases.
Signers are not asking for unrestricted public access to the most cyber-capable models. Their proposed programs would offer early, controlled access to advanced systems, including prerelease versions when appropriate, with enough computing capacity for long-running review tasks. The letter also asks for secure spaces to inspect private or embargoed code and direct communication channels with lab security teams.
The letter asks labs not to limit participation to major companies and governments. Small nonprofits and independent maintainers protect software used by millions of people, the coalition argues. Many lack the institutional ties needed to enter existing programs.
"Defenders need the frontier," the letter concludes. "Please give them a fair head start."
The letter remains open for additional signatures.
Drafted by a large language model from the source reporting linked above, then screened by automated publishing checks. It is not read by a journalist before publication. Some articles cite our Alpha Score. Verify prices and figures against the original source. Educational coverage, not personalized advice.