A coalition of more than 40 cryptocurrency, cybersecurity and open-source organisations is pressing leading artificial intelligence developers to give vetted defenders controlled access to their most powerful models, warning that safeguards designed to prevent cyber abuse may be handicapping legitimate security work. Coinbase, Block, BitGo, Strategy, Blockstream, Galaxy, MARA, Brink, OpenSats, Chaincode Labs, Spiral, Trezor and Unchained are among the organisations supporting the initiative led by the Bitcoin Policy Institute. The open letter, published on August 10, seeks formal trusted-access programmes for specialists protecting Bitcoin and other open-source financial infrastructure.
The campaign reflects a widening concern within the digital-asset industry that offensive uses of frontier AI could advance faster than the tools available to researchers responsible for finding and fixing vulnerabilities. Highly capable models can inspect large software repositories, identify potential weaknesses and automate technically demanding cybersecurity tasks, compressing work that could otherwise require considerable human effort.
The signatories are not asking AI companies to remove cybersecurity restrictions for the public. Instead, they want approved researchers to receive controlled early access to cyber-capable models, sufficient computing capacity for sustained testing, protected environments for examining sensitive or embargoed software and direct communication channels with laboratory security teams. Independent developers and non-profit organisations should also be eligible, rather than access being limited to large companies, the letter argues.
That distinction is central to the dispute. Frontier laboratories have strengthened safeguards because the same capabilities that help researchers detect vulnerabilities can help attackers discover and exploit them. Cybersecurity knowledge is inherently dual-use: instructions useful for patching a flaw can also assist someone attempting to compromise a system.
Bitcoin developers argue that blanket or overly restrictive controls can create an unintended asymmetry. Criminal groups, sophisticated hackers and state-backed actors may use unrestricted open-weight systems, locally operated models or other tools without the safeguards imposed by major commercial AI providers. Legitimate defenders, meanwhile, can encounter refusals when investigations resemble offensive security activity, even when their purpose is vulnerability discovery and remediation.
The issue carries particular importance for cryptocurrency infrastructure because security failures can translate rapidly into irreversible financial losses. Bitcoin wallets, signing devices, cryptographic libraries, node software, exchanges, custody platforms and payment systems form an interconnected ecosystem in which flaws can expose assets or sensitive credentials. The open-source nature of much of that infrastructure allows extensive public scrutiny, but also gives attackers access to the same underlying code.
AI developers have already begun experimenting with models that provide stronger cybersecurity capabilities under additional controls. OpenAI has expanded a Trusted Access for Cyber programme that permits vetted users to conduct authorised defensive work such as vulnerability research, malware analysis, red teaming, threat intelligence and incident response. Its access conditions require users to own, operate or have explicit permission to test the affected systems.
Anthropic has also expanded defensive cybersecurity initiatives. Its Project Glasswing work has included technology designed to find vulnerabilities in widely used software, while selected security teams have been offered access to specialised tools for identifying weaknesses more efficiently. The company has acknowledged that frontier-model cybersecurity knowledge can serve both defensive and harmful purposes, making access controls a continuing technical and policy challenge.
Pressure to refine that balance has increased as frontier models become more autonomous and more capable of completing complex cyber tasks. AI laboratories are attempting to build controls that distinguish authorised research from activity that could materially enable attacks, while security researchers contend that classification systems can struggle to recognise legitimate work involving exploit development or adversarial testing.
The Bitcoin coalition wants trusted-access schemes to become a standing part of frontier-model deployment rather than an exception negotiated after individual researchers encounter restrictions. Its proposal would allow qualified defenders to study emerging AI capabilities before the same techniques become widely available, giving maintainers time to locate vulnerabilities, coordinate disclosures and deploy patches.
The request also places open-source developers more firmly inside the broader debate over frontier AI governance. Research on AI oversight has increasingly emphasised controlled access and independent evaluation as mechanisms for assessing powerful systems without publicly releasing dangerous capabilities. That approach attempts to preserve safeguards while giving specialised researchers enough access to test claims about security, reliability and misuse.
Topics
Cryptocurrency