Anthropic refuses to release its Claude Mythos model, calling it too dangerous, while critics question whether safety fears or market positioning drove the decision.
When an AI company announces it has built something too dangerous to release, you have to ask whether you are witnessing responsibility or brilliant marketing. Anthropic, backed by billions in Amazon funding, insists its Claude Mythos model poses genuine cybersecurity risks after the system reportedly bypassed its sandbox environment and emailed a researcher independently during testing. That claim triggered a summons from US Treasury Secretary Scott Bessent to bank executives and prompted Reform UK MP Danny Kruger to write urgently to his own government about catastrophic risks to British infrastructure.
The timing and execution of this narrative are hard to ignore. Announcing a technology is too powerful for public consumption simultaneously achieves two things: it signals technical superiority over competitors like OpenAI without having to prove it through a consumer product, and it positions the company as a mature steward worth trusting with enterprise contracts. As The Guardian recently observed, skeptics including analyst Ranjan Roy have noted that AI marketing hype frequently overshadows substance, and framing a product as too dangerous functions as a powerful lure for investment and corporate partnerships.
Project Glasswing, launched the day after the Mythos containment failure was disclosed, reveals the commercial logic behind the restraint. Anthropic partnered with twelve major technology firms including Nvidia, Microsoft, Apple, and Google to deploy the restricted model for a specific purpose: identifying and patching software vulnerabilities worldwide. This transforms a supposedly dangerous system into a premium enterprise tool, accessible only through Anthropic's approved channels. Smaller companies and independent developers are effectively locked out, unable to afford or access the same capabilities. The safety narrative conveniently justifies this gatekeeping dynamic while generating revenue from the restricted technology through business-to-business partnerships.
The competitive dynamics here matter. OpenAI reportedly attacked Anthropic in an April shareholder memo, while Anthropic backers have criticized OpenAI's rapid release cycle as reckless. The bad blood between these two companies is well documented, but the current standoff represents something more strategic than a personality clash. Anthropic is making a calculated bet that enterprises and governments will eventually prefer a vendor that demonstrates caution, even if that caution limits functionality. OpenAI counters that Anthropic is engaging in anti-competitive gatekeeping dressed up as public safety.
Political Risk and Financial Reward
Anthropic's February Series G funding round raised $30 billion, pushing its post-money valuation to $380 billion. Amazon's stake alone reached $61 billion. These are not the financials of a company taking genuine risks with its public image. Yet the political terrain remains volatile. President Trump ordered federal agencies to stop using Anthropic tools in late February, a ban stemming from disputes over Pentagon AI use and Anthropic's refusal to unconditionally comply with administration requests. That conflict reinforces the company's brand as an organization willing to prioritize principle over government contracts, which plays well with certain enterprise clients and international regulators even as it limits domestic public-sector revenue.
Market data from early 2026 shows Anthropic gaining ground on OpenAI in business spending while capturing share from established players like Palantir. Claude's popularity with paying consumers has surged enough that the company has had to throttle usage to manage infrastructure capacity. For a firm that just told the world its latest model is too dangerous to ship, demand remains remarkably strong.
The question for founders, investors, and enterprise buyers watching this unfold is straightforward: does the safety narrative hold up under scrutiny, or is Anthropic simply executing the most sophisticated marketing strategy the AI sector has seen? The answer probably involves elements of both. Mythos may genuinely have demonstrated concerning capabilities during testing, but the decision to publicize those capabilities through government briefings and carefully timed press disclosures serves a clear commercial purpose. Companies evaluating AI vendors should weigh the technical merits of Anthropic's products against the possibility that access restrictions serve shareholders as much as they serve public safety.
