Anthropic’s Mythos spots security flaws fast, but burns budgets

Finding software bugs that have hidden in codebases for years is every security team's dream. Finding out it takes $1 million in tokens every few weeks to do it is their CFO's worst nightmare.

That is the exact dilemma Anthropic is presenting to the enterprise world with Mythos, its high-powered AI model currently being tested for cybersecurity tasks.

Millions in tokens, but 20 zero-days down

According to a report from The Information, early enterprise testers are blown away by what Mythos can do. Sam Rubin, a senior vice president heading threat intelligence at Palo Alto Networks, shared that Mythos uncovered more than 20 vulnerabilities in just three weeks. Several of those flaws had gone completely undetected for years.

Security firm Zscaler has also been pointing Mythos at its own internal product code, confirming that while it is vastly more expensive than existing code scanners, the results hold up.

Here is the catch: it chews through compute at a terrifying rate. Palo Alto Networks reportedly burned through about $1 million worth of tokens in a very short window. Another testing firm warned that token bills could easily cross multiple millions of dollars in a matter of weeks.

Anthropic covered testing costs through early subsidies, meaning Palo Alto Networks was not left holding the bag. But those subsidies will not last forever.

Six times the price of Opus

Anthropic has indicated that Mythos will cost roughly six times more per token than Opus, which is already considered one of the priciest frontier models on the market.

Yet several security executives argue the sticker shock is misleading. One British organisation testing the model noted that while token prices are 6x higher, Mythos's sheer efficiency means the actual job cost ends up being only around twice that of Opus.

For enterprise defenders, the calculus is pretty simple. Ryan Downing, CIO at Principal Financial Group, which manages $800 billion in assets, pointed out that the traditional grace period between a flaw being discovered and actively exploited is basically gone. When a single major breach carries hundreds of millions of dollars in lawsuits and business downtime, a million-dollar token tab suddenly looks like a bargain.

Why it matters

We think Mythos signals a dramatic shift where enterprise cybersecurity becomes an outright financial war of token consumption.

For everyday teams and small startups, Mythos is simply not a practical option right now. Burning through millions of dollars in weeks restricts this tier of defensive intelligence to Fortune 500 balance sheets and specialized security vendors.

At the industry level, it sets off an uncomfortable arms race. If offensive actors start wielding AI with this level of vulnerability-finding prowess, the time-to-exploit shrinks toward zero. Companies will have no choice but to continuously run expensive scans just to survive.

Anthropic’s business play here is undeniably sharp: bypassing commodity chatbot pricing to target high-stakes security budgets where millions are pocket change. The real danger, however, is a widening security gap where only the richest platforms can afford AI-grade armor, leaving the rest of the open software ecosystem dangerously exposed.