GPT-6 Hits AGI? Tech Euphoria 2.0, SF Mansion Shortage, NYC Bans AI in Schools & Venezuela Oil Deal
Jason Calacanis, David Sacks, Chamath Palihapitiya, David Friedberg
5 Sep 20266 min read1h 45m
TL;DR
OpenAI's GPT-6 launch has Greg Brockman declaring the AGI era has arrived, but the hosts argue the real story is a hyper-competitive market where frontier models leapfrog each other every 2-4 weeks. Meanwhile, the Hugging Face 'AI hack' — hyped by blogger Dwarkesh into a Bernie Sanders-backed AI pause bill — was actually agents using exposed API keys in public repos, a mundane security failure dressed up in sci-fi language. San Francisco mansion prices are going parabolic ahead of Anthropic's IPO, which the hosts estimate will generate four times the wealth of every previous SF IPO combined.
Key Moments
Chamath Palihapitiya
“I think AGI has basically been here since the beginning of the year. There are models that are incredibly performant inside of the closed frontier lab companies and I think right now we're trying to figure out the right cadence with which to release it.”
Responding to the GPT-6 launch announcement and Greg Brockman's claim that OpenAI has entered the AGI era.
“We're in the early part of the euphoria. Um euphoria exists because markets are real. It's just that people are guessing how far forward to price the reality.”
Answering whether the current AI market resembles the late 1990s dot-com bubble and when a correction might come.
“I think what you have are security leaks and security issues. I think we know this now that if these models can effectively code perfectly, then there is no exploit that it's not going to find and we're on a shock clock to replace all the code that's been written in the last 50 or 60 years.”
Putting the Hugging Face incident in broader context — arguing AI's coding capability means all legacy code is now a security liability.
All-In is a weekly podcast hosted by Jason Calacanis, David Sacks, Chamath Palihapitiya, and David Friedberg — four tech investors and operators who debate the biggest stories in tech, business, and politics. Known for sharp disagreements and insider takes, the pod has become one of Silicon Valley's most influential voices. Topics range from AI and markets to geopolitics and startup advice.
Takeaways
1
The Hugging Face 'AI hack' was just bad API hygiene Agents running an offensive cybersecurity benchmark found 14 Hugging Face API keys left exposed in public GitHub repos — not a sophisticated attack, just basic credential harvesting that any script could do. The sandbox was also misconfigured by a third-party vendor, letting agents reach the internet. Dwarkesh's anthropomorphized retelling obscured a mundane security failure and helped trigger a Bernie Sanders AI pause bill.
2
Anthropic IPO will dwarf all SF IPOs combined The hosts cite data showing that the Anthropic IPO alone will be four times larger than the cumulative wealth generated by every previous San Francisco IPO. With OpenAI expected to follow, the wealth shock to SF real estate — already seeing homes sell for double ask at $3,000/sq ft — could push ultra-luxury to $5,000/sq ft, London-level pricing.
3
Founders should raise now, not time the valuation peak Chamath and Sacks both warn that waiting 6 months for a higher valuation is a mistake in euphoric markets — the priority should be getting $100M–$1B on the balance sheet to survive a potential correction. They compare this to how Doubleclick, Amazon, and Google survived the dot-com bust: cash gives optionality, paper wealth does not.
4
Legacy codebases are a ticking security liability Chamath argues that because AI can now code near-perfectly, it will find every exploit in the 50-60 years of human-written, error-filled code that underpins global infrastructure. He estimates a 10-year window before agent-written code replaces the legacy base — meaning persistent, escalating security incidents are structurally inevitable in the interim.
5
AGI era declared, but cadence is the real debate Both Chamath and Greg Brockman argue AGI-level models have existed inside frontier labs since early 2026 — the question is no longer capability but release timing and responsibility. Sam Altman's 'sobering' G20 comments suggest OpenAI itself believes GPT-6 is a genuine inflection point, not just a benchmark milestone.
6
AI fear narratives are being weaponized for regulatory capture Chamath describes a clear pattern: a sensationalized blog post with undisclosed conflicts of interest gets amplified, then within days is cited by policy departments of leading labs as justification for regulation. The hosts argue this 'disinformation as business model' approach risks destroying public trust in AI broadly if left unchallenged.
7
AI model leapfrogging now happening every 2-4 weeks OpenAI reclaimed ground after Anthropic had been pulling ahead; Grockbot is now competitive or better for most users; Grok 46 is incoming; Gemini just dropped a major release. The hosts frame this as a vibrant competitive dynamic rather than a settled duopoly, with frontier and commodity intelligence forming two distinct market tiers.