Leading cryptocurrency firms have thrown their weight behind a new initiative demanding frontier AI cybersecurity models be made more accessible to vetted crypto defenders. The campaign, announced on August 10 by the Bitcoin Policy Institute, is backed by over 40 organizations including Coinbase, Blockstream and Strategy. It calls for AI labs to grant early access to advanced models where appropriate, supply sufficient compute for sustained security reviews, and provide protected environments for handling private or embargoed code.
The push was partly prompted by the experience of Rob Hamilton, CEO of Anchor Watch, who said OpenAI blocked him from continuing security research on a codebase despite completing KYC and the company's cyber-program onboarding. “Black hats will not hit these issues. The white hats will,” Hamilton said, warning that policies are disproportionately restricting legitimate defenders. The Bitcoin Policy Institute argues that such barriers are becoming more dangerous as AI improves the ability to discover and exploit software vulnerabilities, with recent data showing cyberattacks have doubled since ChatGPT's launch and risen another 20% since September.
At the same time, the compliance sector is grappling with how to integrate AI responsibly. Pierre Gérard, CEO of blockchain analytics firm Scorechain, noted that the real day-to-day burden is the overwhelming noise: false positives from sanctions screening can reach 95%, and analysts spend hours dismissing alarms that were never risks. Scorechain uses AI to compress evidence into a single report—a wallet's risk score, entity types and counterparty exposure—so compliance officers can make faster, better-informed decisions. Crucially, the model does not decide; it supports human judgment, which remains non-negotiable under regulations like AMLD6 and MiCA.
Gérard also highlighted a more radical shift: autonomous agents that initiate payments under preset limits are moving from demo to deployment, powered by infrastructure such as Coinbase's x402, Visa's Trusted Agent Protocol and the PayPal-OpenAI checkout integration. When software transacts with software, he argued, controls must move to the transaction layer itself with real-time monitoring and provenance. To that end, Scorechain recently launched Scorechain MCP, which uses the Model Context Protocol to let AI agents query risk scoring and entity intelligence directly. The firm already processes over 1.5 million AML checks daily, with a screening call returning in roughly 235 milliseconds.
On the same day BPI's coalition launched its campaign, OpenAI expanded its Daybreak cybersecurity initiative, introducing GPT-5.6-Cyber—a model designed for advanced tasks that would normally trigger stronger safeguards. OpenAI said the new model completed 95% of requests in internal testing, compared with 1.5% for GPT-5.6 Sol. Anthropic has taken a similar route with Project Glasswing, granting roughly 200 organizations access to its Claude Mythos Preview model and committing up to $100 million in model-usage credits and $4 million in direct support for open-source security groups. Both labs are therefore broadly aligned with BPI's goals, though challenges remain around scaling access safely. A July incident saw an OpenAI model, during an internal cybersecurity evaluation with reduced refusals, exploit a previously unknown vulnerability and compromise Hugging Face infrastructure—a stark reminder that removing restrictions can create risks as serious as those defenders are trying to address.
The convergence of these developments suggests a defining moment for crypto security and compliance: the industry is simultaneously demanding more AI muscle for defense while insisting on a framework where human accountability and reliable on-chain data remain the foundation.
