Coinbase in talks with Anthropic for Claude Mythos access

Coinbase in Talks with Anthropic for Claude Mythos Access

Coinbase is negotiating restricted access to Anthropic’s Claude Mythos Preview through Project Glasswing to help identify and patch software vulnerabilities.

Coinbase is in talks with Anthropic to obtain restricted access to Claude Mythos Preview under Project Glasswing. The discussions center on using the model to find and remediate software vulnerabilities that could affect the exchange.

Project Glasswing launched in early April 2026 and grants a limited set of partners preview access to Claude Mythos, Anthropic’s frontier security model. Anthropic described the initiative as “an urgent initiative to help secure the world’s most critical software” and said Mythos can detect software vulnerabilities “better than all but the most skilled humans.” During internal testing the model reportedly found thousands of previously unknown zero-day flaws, including a 27-year-old bug in OpenBSD and a 16-year-old flaw in FFmpeg.

Anthropic has provided preview access to founding partners including Amazon Web Services, Apple, Google, JPMorgan Chase, Microsoft and Palo Alto Networks, plus more than 40 other organizations that maintain critical software. The company committed $100 million in compute credits and $4 million to open-source security groups for the program. Anthropic has stated Mythos will not be generally available and that its capabilities are expected to be integrated into future Claude releases with additional safeguards. Post-preview pricing is listed at $25 per million input tokens and $125 per million output tokens.

For Coinbase, talks follow a 2025 insider breach that exposed personal data for about 70,000 users after overseas support agents were bribed. The company refused a $20 million ransom demand and posted a matching bounty for information leading to arrests. Coinbase already uses Claude for customer support operations in more than 100 regions.

Anthropic’s security research has shown that AI agents can autonomously discover and exploit smart contract vulnerabilities. In simulated tests, the company reported AI agents uncovered about $4.6 million in potential exploits, a result that highlights risks to blockchain infrastructure and custodial platforms.

It remains unclear whether Coinbase will secure formal partnership status within Project Glasswing or receive the same level of preview access granted to founding partners. Any agreement would be subject to Anthropic’s stated safeguards and the restricted terms of the Mythos preview program.

The discussions take place amid reports that cryptocurrency exchanges face increasingly sophisticated threats that use automated tools and large language models to probe software, steal credentials and exploit contract code. Anthropic has emphasized controlled access and plans to fold Mythos capabilities into guarded product versions rather than distribute the model publicly.

Content on BlockPort is provided for informational purposes only and does not constitute financial guidance.
We strive to ensure the accuracy and relevance of the information we share, but we do not guarantee that all content is complete, error-free, or up to date. BlockPort disclaims any liability for losses, mistakes, or actions taken based on the material found on this site.
Always conduct your own research before making financial decisions and consider consulting with a licensed advisor.
For further details, please review our Terms of Use, Privacy Policy, and Disclaimer.

Articles by this author

This site is registered on wpml.org as a development site. Switch to a production site key to remove this banner.