Key Highlights
- Discussions are underway between Coinbase and Anthropic regarding access to the Claude Mythos Preview frontier AI model.
- The exchange aims to enhance security infrastructure and boost operational performance through this technology.
- Project Glasswing, launched by Anthropic in early April 2026, provides select organizations with controlled Mythos access.
- The AI model discovered thousands of previously undetected zero-day security flaws during testing phases, including a vulnerability in OpenBSD that existed for 27 years.
- Initial Glasswing partners comprise AWS, Apple, Google, JPMorgan Chase, Microsoft, and Palo Alto Networks.
Coinbase has initiated conversations with Anthropic regarding potential access to Claude Mythos Preview, a specialized AI system designed for sophisticated cybersecurity applications, The Information reports.
This strategic move reflects mounting challenges cryptocurrency platforms encounter as artificial intelligence-enhanced attack vectors evolve beyond traditional defensive capabilities.
Anthropic introduced Project Glasswing during early April 2026. Through this initiative, a carefully selected cohort of organizations receives Mythos access exclusively for protective security applications.
The inaugural partner roster features Amazon Web Services, Apple, Google, JPMorgan Chase, Microsoft, and Palo Alto Networks. Over 40 additional entities responsible for essential software infrastructure also gained entry to the program.
Anthropic allocated $100 million in computational resources alongside $4 million directed toward open-source security communities to support the project.
Throughout testing procedures, Mythos uncovered thousands of zero-day vulnerabilities that had escaped prior detection. Notable discoveries included a 27-year-old security weakness in OpenBSD and a 16-year-old defect within FFmpeg.
These revelations have drawn interest from major platforms such as Coinbase, which currently deploys Claude for customer assistance operations spanning more than 100 geographic markets.
The Strategic Value for Coinbase
Coinbase experienced a significant internal security incident in 2025. Criminal organizations bribed international support personnel, resulting in compromised personal information affecting approximately 70,000 users.
The platform declined a $20 million extortion payment and instead established an equivalent reward for intelligence enabling perpetrator apprehension.
Research conducted by Anthropic has demonstrated that AI agents possess the capability to independently identify and exploit smart contract weaknesses, creating millions in theoretical stolen assets during simulation scenarios.
Such findings position Mythos access as a critical infrastructure component for platforms managing substantial financial transactions.
Pricing Structure and Availability Framework
Mythos will remain unavailable for general commercial use. Anthropic intends to integrate its advanced features into subsequent Claude versions equipped with enhanced protective measures.
Following the preview phase, pricing stands at $25 per million input tokens and $125 per million output tokens.
Confirmation regarding Coinbase obtaining official Glasswing partnership designation or expanded model access remains pending.
Negotiations continue between both organizations, with neither company issuing formal confirmation of any finalized arrangement.
