What happens when your best AI product isn’t a product at all—because it’s busy hacking the world’s software?
Anthropic’s Mythos preview dropped this week, but don’t get excited. They won’t let you touch it. Not you, not me, not even most of their paying customers. After last week’s leaks turned into this week’s headlines, the company confirmed: Mythos found thousands of unknown zero-day vulnerabilities across every major OS and browser. Public release? Nope. Too risky, they say.
“Anthropic determined that public release would ‘significantly amplify cybersecurity risks.’”
That’s straight from their announcement. Unsettling, right? But here’s my first gut check after 20 years chasing Valley hype: who’s really scared here?
Why Did Anthropic Lock Down Mythos?
Look, we’ve heard the safety sermons before—endless debates, theoretical what-ifs. Anthropic skipped the talk. They acted. Released it via Project Glasswing, whitelist-only to 11 big shots: AWS, Apple, Cisco, CrowdStrike, Google, JPMorgan, Microsoft, NVIDIA, Palo Alto, Broadcom, Linux Foundation. That’s not openness; that’s a VIP club for the usual suspects.
They tossed in $100M credits and $4M to open-source security outfits as a peace offering. Noble? Maybe. But criteria for those partners? Crickets. And who decides ‘too dangerous’? Anthropic, for now. Smells like the start of AI gatekeeping, where labs play god with tech that could rewrite cybersecurity.
Short version: this ain’t sharing. It’s controlled detonation.
My unique angle—and you’ll not find this in the press releases—echoes the Manhattan Project. Back then, the U.S. built the bomb but hoarded it, doling access to allies only. Anthropic’s doing the same with digital nukes. Partners get the boom; everyone else gets fallout warnings. Bold prediction: by 2027, we’ll see ‘Mythos 2.0’ licensed like export-controlled weapons, with governments knocking.
Has Anthropic’s Revenue Really Blown Past OpenAI?
While Mythos hides in the vault, Anthropic’s wallet exploded. Bloomberg pegs annual recurring revenue at over $30B—up from $9B end of 2025. That’s a 3.3x leap in months. Over 1,000 enterprise customers dropping $1M+ yearly. They inked 3.5 gigawatts of TPU compute from Google and Broadcom, deliveries in 2027. Broadcom’s even banking $21B from Anthropic this year, doubling next.
Numbers like that? Unheard of in software. Caveat: mixes subs with cloud prepays, so grain of salt. Still, Anthropic’s now the cash king, dethroning OpenAI. But who profits most? Those chip giants supplying the juice. Compute’s the new oil; models are just excuses to burn it.
And the markets? Freaked. Mythos’s vuln-hunting prowess tanked the S&P Software index 2.6% in a day—YTD down 25.5%. Cyber stocks like CrowdStrike, Cloudflare plunged 5-6%. Investor math: if one AI shreds your moats in minutes, what’s left? Legacy software’s toast.
But wait—AI-native Sec-DevOps tools? They’re the phoenix. Patching zero-days in real-time becomes the moat. Watch those bets shift.
This triple shock sequel proves the leaks were tame. Anthropic’s juggling ‘safety saint’ and ‘profit beast.’ Tension that’ll define 2026.
OpenAI’s $122B Mega-Round: Diversifying or Desperate?
Meanwhile, OpenAI closed the fattest private round ever: $122B at $852B valuation. Amazon drops $50B—now their top sugar daddy. NVIDIA, SoftBank $30B each, banks the rest. Microsoft’s grip slips; Stargate drama to this Amazon play screams diversification.
CFO Sarah Friar teases late-2026 IPO with retail shares. Codex hits 3 million what? Article cuts off, but point stands: OpenAI’s chasing capital ceilings while Anthropic redefines technical ones.
Here’s the cynicism: these moves scream ‘compute arms race.’ Not model strength, but governance, trust, cash. Who wins? Hyperscalers and chipmakers. End users? Pray the safety nets hold.
Software stocks crashed because investors finally asked my perennial question: who’s actually making money here? Turns out, not the vuln-plagued middlemen.
The AI race pivoted hard. Forget ‘strongest model.’ It’s ‘who controls the keys—and the dollars.’ Anthropic leads on both, for now.
🧬 Related Insights
- Read more: Microsoft’s Stealth Service Mesh Erases Complexity for Good
- Read more: Nine Markdown Files That Reign in Rogue AI Coders
Frequently Asked Questions
What is Anthropic’s Mythos model?
Mythos is Anthropic’s most advanced AI, capable of autonomously finding thousands of zero-day vulnerabilities in major OSes and browsers. They won’t release it publicly due to cyber risks; access limited to 11 vetted partners via Project Glasswing.
Did Anthropic surpass OpenAI in revenue?
Yes, reportedly over $30B ARR now vs. OpenAI, up from $9B in four months. Includes enterprise deals and compute prepays—stunning growth, but verify the breakdowns.
Why did software stocks crash after Mythos news?
Fear that AI like Mythos erodes software moats by spotting flaws instantly, hitting cyber firms hardest. Shifts investment to AI patching tools.
