Mythos false positive rate: 30%
Claude's Mythos model had a 30% false positive rate when used for vulnerability detection — great for offense, but problematic for defense.
Snapshot · All-In with Chamath, Jason, Sacks & Friedberg
Nikesh Arora: Mythos is Real, Analytical SaaS is Dead, and Google can be a $10T company
Explore episode
Jun 8, 2026
Where this was said
If models become a utility, where will the money be made?
At 19:08 · chapter starts 14:06
The conversation turns to model economics and where value will ultimately accrue in the AI stack. Nikesh offers a vision of AI models becoming pure utilities — you buy 120 IQ for routine tasks, 250 IQ for complex ones, paying fractions of cents versus dollars. The implication: model providers won't own the profit pools. That's why OpenAI is racing to build Codex and Anthropic is building Claude Code — they're attacking the application layer, which is where enterprise software spending already lives. [1] — Nikesh Arora "AI models will commoditize into a utility layer where you buy intelligence on demand at different price-performance tiers. The profit pools…" 14:06 But here's Nikesh's key observation: the AI-native application companies that will replace legacy SaaS are not yet fully formed. Fifty thousand enterprises need the same HR system, the same sales platform — it's wildly inefficient for each to build their own on top of raw model APIs. The opportunity waiting to be captured is building that next generation of enterprise application companies, complete with harnesses, memory, and data pipelines.
Models Are Becoming a Utility — Profit Is in Applications
AI models will commoditize into a utility layer where you buy intelligence on demand at different price-performance tiers. The profit pools are in the application layer — which is why OpenAI is pushing Codex and Anthropic is pushing Claude Code. They know that's where the money will be.
Entire Frontier Model Weights Fit on a USB Stick
A major AI model CEO told Nikesh Arora that the full weights of their newest frontier model fit on a single USB stick and can be distilled in under 48 hours. That fact alone makes a 6-month export control regime essentially meaningless — the IP walks out the door.
Model weights fit on a USB stick
A CEO of a leading AI model company told Nikesh Arora that the entire weights of their newest frontier model fit on a single USB stick, distillable in under 48 hours.
The False Positive Problem Is AI's Real Enterprise Blocker
Mythos had a 30% false positive rate — meaning nearly 1 in 3 vulnerabilities it found didn't exist. For offense that's fine; for defense it's catastrophic. The real challenge of deploying AI in enterprise isn't getting the newest model, it's driving false positives from 20% down to 0.01%.
Similar snapshots
MVP built in one week
Sam's initial MVP was coded in approximately one week using ChatGPT voice mode and copy-pasting code, with no prior technical experience.
Discord vs email for young users
Sam argues Discord is 10x better than email for building relationships with younger users who rarely check their inbox.
Monthly infra costs ~$780–$980
Sam's monthly operating costs include Cursor ($200), AI image generation ($100), AI video generation ($200), hosting ($100), email marketing ($80), and AI compute ($300–$500).
ChatGPT-powered pain point research
Sam recommends copying days of Discord chat history into ChatGPT and prompting it to list recurring pain points as a fast, free market research technique.
50 free tools built total
Bhanu and his team built approximately 50 free tools to attract search traffic, each linked back to SiteGPT.
New tool built in under 5 minutes
With AI coding tools like Cursor, Bhanu can now create a new free marketing tool in less than 5 minutes by referencing existing tools.
KD filter < 10 for rankable keywords
Bhanu filters Ahrefs keyword results to show only those with a keyword difficulty below 10, making them realistic ranking targets for any decent website.
Minimum 1,000 monthly search volume
Bhanu sets a minimum search volume of 1,000 monthly searches when selecting keywords to target with free tools.
20 downloads/day at launch
PropGPT averaged 20 downloads per day right after launching on the App Store through influencer marketing.
