A coalition of prominent bitcoin and cryptocurrency companies is pushing back against safety restrictions on frontier artificial intelligence models, arguing that the same tools are freely available to malicious actors but denied to those trying to protect the ecosystem. In an open letter organized by the Bitcoin Policy Institute, more than three dozen firms, including Coinbase, Block, BitGo, and ARK Invest, urged major AI laboratories to give open-source security researchers early access to their most advanced systems.
The letter, which was published this week, claims that safety guardrails on frontier models are inadvertently creating a double standard. While cybercriminals and hostile nation-states can access these models through less regulated channels, legitimate security professionals face throttled capabilities, content filters, and limited access to the underlying technology. This asymmetry, the signatories argue, is leaving critical infrastructure more vulnerable than ever.
Defenders locked out, attackers unshackled
The core grievance is straightforward: the developers who maintain bitcoin’s core software, peer-to-peer payment processors, and Lightning Network nodes are unable to use the full power of leading AI models to audit code, simulate attack vectors, and design defenses. At the same time, attackers are using the same models without ethical constraints, often exploiting open-source versions that lack safety filters or turning to underground APIs.
According to the letter, “Bitcoin Core developers are locked out of trusted-partner programs and hampered by safety filters on public models.” This means that the very people working to secure one of the world’s most attacked digital asset networks cannot access the tools that would let them keep pace with evolving threats. The signatories are asking AI labs to provide early access to frontier models, compute resources, secure testing environments, and direct communication channels with lab security teams.
The demand is not a casual suggestion but a formal appeal signed by influential players in the digital asset industry. The Bitcoin Policy Institute, a non-profit research and advocacy group, organized the effort. It was joined by major exchanges, wallet providers, and investment firms, signaling broad industry concern about the growing security gap.
Real-world exploits highlight the risk
The letter cites recent security incidents as evidence that the current approach is failing. In particular, it references exploits against BTCPay Server and Lightning nodes — two critical pieces of infrastructure for bitcoin merchants and users. Those vulnerabilities were uncovered, in part, by a volunteer “Bitcoin Red Team” that used AI tools to probe for weaknesses in open-source software.
The Bitcoin Red Team is an informal collective of security researchers who use machine learning to identify code vulnerabilities, test edge cases, and simulate adversarial scenarios. Their work has already produced concrete results, but it has been conducted with the same models that are criticized as too restrictive. The letter argues that a dedicated, properly funded collaboration with AI labs would accelerate these efforts and catch bugs before they are exploited in the wild.
The broader context is troubling for bitcoin’s security posture. Bitcoin is a high-value target, and the sophistication of attacks has increased sharply in recent years. From exchange hacks to clipboard malware to elaborate social engineering campaigns, threat actors are constantly developing new methods. The rise of generative AI has only intensified the arms race, as attackers can now automate phishing, analyze code for weaknesses, and craft malware at scale. Defenders are expected to counteract these threats with manual audits and limited access to the same technology.
Why AI labs are cautious
Major AI labs, including OpenAI, Anthropic, and Google DeepMind, have implemented safety guardrails on their most powerful models. These restrictions are designed to prevent misuse, particularly in areas like cyberattacks, bioterrorism, and disinformation. Early access programs, often referred to as trusted-partner or red-team initiatives, are typically limited to a small number of vetted organizations.
AI labs argue that granting broad access to open-source security researchers could increase the risk of misuse. Models that can analyze code for vulnerabilities are equally useful to attackers looking for exploits. Allowing more users, even well-intentioned ones, increases the chance that model outputs will be leaked or repurposed for malicious ends.
However, bitcoin firms counter that the current approach is naive. Attackers have already demonstrated the ability to fine-tune open-source models or use jailbroken versions of commercial models to achieve their goals. In many cases, they do not need early access because they can simply use uncensored models available on platforms like Hugging Face. Meanwhile, defenders—who must comply with corporate policies and ethical guidelines—are left to work with both hands tied behind their backs.
Asymmetric warfare in the AI era
The letter’s central point is about asymmetry. In cybersecurity, defenders must be right 100% of the time, while an attacker only needs to be right once. AI has the potential to level this playing field, but only if defenders have access to the most capable models. Without such access, the gap widens, and the consequences for bitcoin users could be severe.
Bitcoin’s decentralized nature means that the burden of security falls on many participants. Independent developers, node operators, and wallet providers all contribute to the overall resilience of the network. The ecosystem has historically relied on open-source peer review to catch bugs, but the volume and complexity of modern code have outpaced manual inspection. AI-assisted auditing is increasingly seen as a necessary tool to keep up.
The signatories are not asking for unrestricted public access to frontier models. Their request is more targeted: early access for vetted security researchers, based on clear criteria and under conditions that prevent misuse. They point to existing trusted-partner programs as a model, but note that the current gatekeeping is too narrow and excludes many people working on open-source projects like Bitcoin.
Specific demands laid out
The letter lists several concrete requests. First and foremost, AI labs should include open-source security researchers in their early-access programs for frontier models. This includes not just the models themselves, but also the computational resources needed to run them for large-scale auditing tasks. Second, labs should provide secure testing environments where researchers can work with sensitive code and data without fear of exposure. Third, there should be direct communication channels between the labs and the security community, so that threat intelligence can be shared in real time.
Finally, the letter calls for a reconsideration of safety filters that block legitimate cyber-defense tasks. For example, a model might refuse to generate a proof-of-concept exploit for a known vulnerability, even when the researcher’s intent is to demonstrate the bug to a software developer so it can be patched. These blanket refusals hinder defensive work while doing little to stop attackers, who can easily repurpose malicious output from other sources.
The coalition also expressed willingness to work with AI labs to develop appropriate safeguards. The goal is not to eliminate safety measures but to create a more nuanced system that distinguishes between offensive misuse and defensive research. The signatories argue that a partnership model, rather than an adversarial one, would produce better outcomes for both industries.
Industry response and next steps
The letter has already generated discussion within the AI and crypto communities. Some AI researchers have expressed sympathy with the argument, acknowledging that overly broad guardrails can impede security work. Others remain skeptical, pointing out that even well-intentioned researchers can make mistakes, and that the risk of accidental release of dangerous capabilities is real.
For now, no major AI lab has publicly responded to the letter. However, some recent developments suggest a shift in thinking. Anthropic, for instance, launched Project Glasswing, a security initiative that focuses on defending AI systems themselves. Kraken’s parent company, Payward, recently joined the project, signaling a desire for closer collaboration between crypto and AI security teams. While Project Glasswing is primarily about protecting AI systems, it could pave the way for more joint efforts.
The bitcoin industry is no stranger to political and regulatory battles. This letter is not the first time that bitcoin firms have called for better treatment from technology providers, and it likely will not be the last. But the issue at hand is more fundamental than mere access to APIs. It is about whether the people responsible for securing critical financial infrastructure will be equipped to face the threats of the next decade.
As AI continues to evolve, the line between offensive and defensive use will become increasingly blurred. The bitcoin firms’ letter is an early attempt to shape the access policies of AI labs before that line hardens into permanent rules. Whether the labs will listen remains to be seen, but the conversation has begun.
Source: Coindesk News