Wall Street forgot that software tokens cost real money. For the past two months, tech investors acted like Anthropic's new Mythos model was a magic wand that would print infinite growth for the cybersecurity sector. It wasn't entirely crazy to think so. When Anthropic dropped Claude Mythos Preview, it felt like a watershed moment. The AI demonstrated an eerie ability to autonomously find and exploit zero-day vulnerabilities, including a 27-year-old flaw in OpenBSD.
Cybersecurity stocks went parabolic. CrowdStrike surged 60% in May alone. Zscaler and Palo Alto Networks rode the exact same wave.
Then came the first week of June, earnings reports dropped, and the music stopped.
CrowdStrike dropped 9% in a single day despite beating quarterly estimates. Zscaler issued a weak fourth-quarter revenue forecast that triggered immediate downgrades. The brutal reality of the software business caught up to the AI hype cycle. High-flying valuations ran directly into an expensive, resource-heavy tech deployment, proving that the Mythos boom is going to look a lot different than the market anticipated.
The Brutal Physics of Six-Times Token Costs
Investors missed a massive variable in their spreadsheets. Mythos isn't just more capable than previous models like Claude Opus; it is blindingly expensive. Anthropic is charging roughly six times more per token for Mythos than it does for Opus.
That changes the unit economics of security operations instantly.
Early testing reports show that Palo Alto Networks burned through roughly $1 million worth of tokens in a matter of weeks during its initial trials. Right now, Anthropic is heavily subsidizing these costs through $100 million in usage credits allocated to its Project Glasswing partners. But those credits will dry up. When they do, enterprise buyers will have to foot the bill.
Some analysts argue that the efficiency gains compensate for the premium. A British research organization noted that while token prices are six times higher, the speed of autonomous testing means customers might only see a 2x increase in actual operational costs relative to output. Even Zscaler CEO Jay Chaudhry admitted the tool adds fuel to the fire, acknowledging it is expensive but arguing it remains worth the investment.
But corporations operate on fixed budgets. Chief Information Officers cannot simply manifest an extra $5 million a year out of thin air to feed an AI engine. We're already seeing the friction. Buyers are hesitating, net new customer growth is ticking downward for legacy vendors, and the massive revenue windfall that Wall Street modeled for this quarter didn't materialize.
The Disruption Inside Project Glasswing
The threat to traditional software vendors isn't just external. It is happening from within the very partnership programs designed to test this technology.
Anthropic just announced a massive expansion of Project Glasswing, pushing the model out to 150 organizations across 15 countries. It's no longer a tight US-UK research experiment. Heavyweights like Okta, Samsung, SK Hynix, Euroclear, and the New York Stock Exchange are getting direct access.
Take a look at what happens when an enterprise deploys an autonomous reasoning engine like Mythos. It doesn't complement legacy pattern-matching vulnerability scanners. It replaces them.
The initial cohort of 50 partners used the preview model to flag more than ten thousand high- or critical-severity vulnerabilities. If a company can deploy an AI agent to crawl its codebase, find a zero-day flaw, and write the patch autonomously, why would it continue to pay millions of dollars annually for traditional application security dashboards?
This is a structural shift. Legacy vendors are caught in a vise. They must either pay massive token fees to integrate Mythos into their own products, compressing their profit margins, or watch their enterprise clients bypass them entirely to work with frontier AI labs.
Government Secrecy and the Two-Sided Weapon
While public markets freak out over enterprise software margins, the geopolitical landscape reveals why this tech is valued so highly. Anthropic recently filed for an initial public offering targeting a $1 trillion valuation. You don't pitch a trillion-dollar valuation based on a code scanner. You pitch it based on critical national infrastructure.
Reports surfaced that Anthropic has embedded forward-deployed engineers directly inside the US National Security Agency (NSA) to help deploy Mythos for offensive cyber operations.
"The best way to build a good defence is to build a good attack," noted a source close to the deployment. "If Mythos is not used to build attack agents, adversaries will find a way to do it."
The UK's AI Security Institute tested the model and confirmed it can execute complex, multi-stage network attacks that would take human teams days to organize. This dual-use reality introduces massive regulatory risk that tech investors are ignoring. The White House is actively debating executive orders to restrict access to these specific models.
If the US government decides to lock down Mythos-class models under strict export controls or defense classification, the commercial TAM (Total Addressable Market) for these cybersecurity companies shrinks overnight. You can't build a sustainable commercial SaaS growth story around a piece of software that the Pentagon might classify as a munition next month.
Your Strategic Next Moves
If you are managing an enterprise security budget or adjusting a tech investment portfolio, stop trading the daily momentum and focus on structural realities.
- Audit Your Security Vendor Stack: Evaluate how much you pay for static pattern-matching scanning tools. If your primary vendors aren't showing a clear, economically viable integration path for reasoning-based AI, prepare to migrate budgets toward direct AI infrastructure.
- Model the Token Burn Rate: If you plan to apply for Project Glasswing or adopt Mythos-class models as they hit wider commercial availability, do not look at the pilot costs. Model your budget around the post-subsidy price of six times Opus token pricing to avoid a sudden cash crunch.
- Watch the Enterprise SaaS Margins: Keep a close eye on the gross margins of pure-play cybersecurity firms over the next two quarters. The companies that successfully pass AI token costs onto their end customers without breaking retention rates are the ones that will survive the post-rally correction.
