Dozens of cryptocurrency and security firms, including Coinbase, Block and BitGo, have signed an open letter urging leading artificial intelligence labs to loosen the safety restrictions on their frontier models, arguing that current guardrails hinder legitimate defensive security work while doing nothing to stop malicious actors.
The Core Complaint
The signatories contend that AI safety filters designed to prevent misuse are inadvertently penalizing the wrong people. When security researchers ask advanced models for help analyzing malware, probing vulnerabilities or simulating attacks, they frequently hit refusals baked into the models to prevent harmful outputs.
Attackers, by contrast, operate outside these constraints entirely. Bad actors can rely on unrestricted or jailbroken models, custom-built tools and offline systems that carry no ethical guardrails whatsoever, leaving defenders working with one hand tied behind their back.
Defenders are being blocked from the very capabilities their adversaries already wield without restriction.
The firms argue this creates a dangerous asymmetry in the ongoing battle to protect digital assets, exchanges and blockchain infrastructure from increasingly sophisticated threats.
What the Firms Are Asking For
The letter calls on AI developers to create pathways that let vetted security professionals access the fuller capabilities of frontier models for defensive purposes. Rather than blanket refusals, the signatories favor a more nuanced approach that distinguishes between legitimate research and genuine abuse.
Key requests from the coalition include:
- Tailored access for verified security teams and researchers
- Clearer policies on what constitutes acceptable defensive use
- Recognition that overly broad safety filters can weaken overall security
For crypto companies in particular, the stakes are high. The industry remains a frequent target of hacks, phishing campaigns and exploits, and firms say they need every available tool to keep pace with adversaries who face no such limitations.
The debate highlights a broader tension in AI governance: how to balance the prevention of misuse against the legitimate needs of the professionals working to defend systems from attack.