Anthropic faces trust challenges as Mythos AI rollout begins

Anthropic’s newest AI model can autonomously detect cybersecurity vulnerabilities and execute complex exploits. And someone already breached the program designed to keep it contained.

Claude Mythos Preview, which launched around April 7-8, 2026, is being called Anthropic’s most advanced model to date. Its cybersecurity capabilities reportedly saturated or exceeded established benchmarks during external evaluations. Anthropic opted for a controlled rollout through a program called Project Glasswing, limiting access to roughly 40-50 vetted organizations focused on defending critical infrastructure.

That plan hit a wall on April 21-22, when reports surfaced of unauthorized access through a third-party vendor breach.

Project Glasswing and the controlled release strategy

The company committed up to $100 million in usage credits for security initiatives through the program. On top of that, Anthropic allocated $4 million in direct donations for open-source security efforts.

External evaluators, including METR and the UK AI Security Institute, assessed Claude Mythos ahead of the limited release. Their findings were notable. The model’s cyber capabilities reportedly exceeded many established benchmarks.

The breach and its fallout

The unauthorized access incident on April 21-22 didn’t come from a direct attack on Anthropic’s systems. It came through a third-party vendor. The breach has drawn scrutiny from governments, cybersecurity experts, and the broader tech community.

OpenAI’s Sam Altman weighed in with remarks suggesting that Anthropic’s marketing strategy may be influenced by fear over the technology’s implications, implying the controlled rollout is partly a PR play designed to make Claude Mythos seem more dangerous, and therefore more impressive, than it actually is.

External evaluators flagged the dual-use nature of the technology directly: a model that can find vulnerabilities can also exploit them.

What this means for the crypto industry

Claude Mythos’s capabilities are prompting a broader rethinking of what “security” means for digital asset platforms, extending beyond traditional smart contract audits to infrastructure components like key management systems and oracles.

Prior leaks of AI models have briefly influenced Bitcoin prices, though no direct connection between Claude Mythos and specific cryptocurrencies has been established so far.

The $100 million in usage credits Anthropic committed through Project Glasswing signals that serious money is flowing into defensive AI applications. But it also creates a new dependency: crypto platforms relying on the same AI company’s tools for defense that other actors might eventually use for offense.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.