Anthropic’s hypocrisy: “we won’t remove safety guardrails for the US government, but we will grant access to our upcoming next-gen Mythos model only to the banks and corporations”

Anthropic, the AI safety-focused company behind Claude, is making waves with its restrictive rollout of a powerful new model called Mythos. The company has stated it will not remove safety guardrails for the US government, yet is granting exclusive early access to its upcoming next-generation system to a curated list of major banks and corporations. This selective access policy highlights a complex prioritization where commercial and critical infrastructure partners are deemed suitable first recipients over sovereign entities, raising immediate questions about the criteria and ethics of such gatekeeping.

The Mythos model itself is described as a compute-intensive system optimized for complex logic and deep technical reasoning. While general-purpose, its reported "emergent" talent for discovering software flaws is a key factor in the current lockdown. As of the reported timeline of April 2026, access is limited to launch partners and vetted organizations. This elite group includes Big Tech cloud providers (Google's Vertex AI, Microsoft's Azure, Amazon's AWS), cybersecurity giants like CrowdStrike and Palo Alto Networks, infrastructure leaders such as Cisco and NVIDIA, and major financial institutions including JPMorgan Chase and a select group of British banks. The inclusion of UK banks followed specific government concerns about financial system resiliency, indicating access is granted based on perceived systemic risk mitigation, not purely commercial terms.

This strategy creates a stark dichotomy: Anthropic is withholding unfettered access from the US government on safety grounds, while simultaneously empowering massive corporations with a tool noted for its flaw-finding capabilities. The move positions Anthropic not just as a model developer, but as a gatekeeper of advanced AI capabilities, deciding which sectors—finance, tech infrastructure, cybersecurity—are "responsible" enough for early adoption. It challenges the narrative of open or broadly democratized AI advancement, instead suggesting a future where the most powerful AI tools are first deployed to fortify the digital and financial architectures of the global economy under the supervision of their corporate stewards.