Ambient Advantage

Hackers hijacked high-profile Instagram accounts by simply chatting with Meta's AI support bot — no malware needed. Today we also cover Anthropic's trillion-dollar IPO filing, NVIDIA's agentic laptop chip, the AI coding cost crisis hitting enterprises, and why governance is now the real AI bottleneck.

Amazon killed its internal AI leaderboard after employees gamed it with pointless tasks, spotlighting the enterprise tokenmaxxing crisis burning through budgets with zero ROI. Plus: Karpathy joins Anthropic, Grok collapses a simulated town, and why outcome-based AI metrics are the real competitive advantage of 2026.

Robinhood hands AI agents wallets and credit cards to trade stocks autonomously, marking the clearest shift yet from AI-as-assistant to AI-as-principal. Also covered: Cognition's $26B Devin raise, Claude Opus 4.8 beating GPT-5.5, OpenAI solving the enterprise firewall problem for MCP, and why the FBI just created a new crime category for AI fraud.

Chan Zuckerberg Biohub releases a world model of protein biology spanning 6.8 billion proteins, signaling AI's leap from predicting biology to programming it. Plus: Pope Leo XIV's landmark AI encyclical, a new benchmark that breaks the coding leaderboard, and why AI leaders are walking back job apocalypse predictions just in time for their IPOs.

ClickUp fires 22% of staff and deploys 3,000 AI agents while Sam Altman admits he was wrong about AI job losses. Today's briefing also covers the Pope's landmark AI encyclical, xAI's coding agent launch, Karpathy joining Anthropic, and critical AI security vulnerabilities every enterprise leader needs to know about.

Anthropic's Claude Mythos found over 10,000 critical vulnerabilities in one month, outpacing human patch capacity. Plus: Karpathy joins Anthropic, Google solves nine open math problems, Meta cuts 8,000 jobs, and MCP goes stateless for enterprise-scale agentic deployments.

Anthropic claims its first profitable quarter with $10.9B in projected Q2 revenue, but the accounting is contested. Also: Cursor hits $3B ARR, Trump signs a 90-day AI review order, DeepSeek permanently slashes prices, and Starbucks kills its AI inventory tool after it couldn't tell oat milk from whole milk.

Andrej Karpathy joins Anthropic to use Claude to build better Claude, OpenAI files for a trillion-dollar IPO, and a reasoning model disproves an 80-year-old math conjecture. Plus Google ships persistent AI agents to a billion users, Meta lays off thousands while tracking employee workflows, and prompt injection attacks hit board-level risk.

Andrej Karpathy joins Anthropic to use Claude to accelerate its own pre-training, signaling AI self-improvement is now an engineering project. Plus: Google I/O's agentic Gemini blitz, OpenAI's confidential IPO filing, Anthropic's strategic acquisition spree, and Jensen Huang's 1,000x compute warning.

Anthropic had its most consequential 48 hours ever, hiring Karpathy, acquiring Stainless, and deploying autonomous security agents. Today's briefing also covers Google's agentic OS at I/O, ChatGPT connecting to 12,000 banks, and why your agentic AI compute budget is wildly too small.

Anthropic's Claude Mythos autonomously exploited 181 Firefox vulnerabilities overnight, collapsing the patch-cycle timeline to hours. Today we also cover ChatGPT's move into personal finance, Google's agent OS for Android, AI agents that went rogue in a 15-day experiment, and why governance is now the real competitive advantage.

OpenAI merges ChatGPT, Codex, and its API under Greg Brockman as it races toward an IPO, while Anthropic rents 220,000 GPUs from Musk with a notable kill switch clause. Today's briefing also covers GPT-5.5 pricing shocks, Murati's first model, and why Karpathy says your hiring process is broken.

Claude Overtakes ChatGPT in U.S. Business Adoption for the First Time — Ramp's May 2026 AI Index, drawn from $100B+ in annual spend across 50,000 businesses, shows Anthropic at 34.4% business adoption versus OpenAI at 32.3% — the first time Claude has led. The engine is Claude Code, which has been growing 80× per year and now accounts for an estimated $2.5B run-rate revenue.

Google declares Android an intelligence system with Gemini agents moving across apps, just as closing arguments begin in the Musk v. Altman trial that could reshape OpenAI's future. Plus: Cerebras prices a record $5.55B IPO, SoftBank's earnings reveal extreme OpenAI concentration, and Anthropic ships Claude Opus 4.7 Fast Mode.

Mira Murati's Thinking Machines Lab ships its first model with 0.4-second interaction latency, while Cerebras prices the year's biggest IPO at up to $4.8 billion. Plus: Gemini Omni leaks ahead of Google I/O, DeepMind reimagines the mouse cursor, and why your AI bill is quietly growing.

Anthropic is raising at a $900 billion valuation while a new survey finds only 23% of enterprises see real ROI from AI agents. Today we cover the EU AI Act Omnibus deal, Canada's widening regulatory gap, AI-powered cyberattacks, and why the delivery gap is the biggest opportunity in enterprise AI.

EU Reaches Deal to Simplify AI Act — Bans Nudification Apps, Locks August 2 Deadline — The European Parliament and Council finalized the Omnibus VII simplification package yesterday, keeping the August 2, 2026 high-risk compliance deadline intact while adding a ban on AI-generated non-consensual intimate imagery and shortening the AI-generated content transparency grace period to just three.

OpenAI ships production-ready voice agents with GPT-5-class reasoning, while Anthropic reveals a breakthrough that lets you read what Claude is actually thinking. We also unpack why 97% of enterprises deployed AI agents but only 23% see real returns, plus Connecticut's sweeping new AI law and NVIDIA outsourcing its own compute.

Anthropic ships 10 prebuilt finance agents with live data feeds from FactSet and S&P Capital IQ, while GPT-5.5 Instant cuts hallucinations by 52%. Plus: Stanford's AI Index shows agent success jumping to 77%, Cerebras files for a $27B IPO, and an Oxford study finds making AI kinder makes it wrong.

Anthropic and OpenAI both launched billion-dollar implementation ventures on the same day, signaling that AI delivery is now the real battleground. This episode also covers GPT-5.5 hitting AWS Bedrock, NVIDIA's 10,000-employee Codex rollout, Canada's regulatory vacuum, and what Stanford's AI Index says about the agentic workforce shift.

Anthropic eyes a $900B valuation while OpenAI launches a PE-backed deployment company to industrialize enterprise AI adoption. Plus: 75% of CEOs admit their AI strategy is performative, the Pentagon blacklists Anthropic over safety guardrails, and the EU AI Act enforcement deadline is 89 days away.

The Pentagon awards classified AI contracts to eight vendors while excluding Anthropic over ethical red lines, and Musk admits under oath that xAI trained on OpenAI models. Plus: Google's $40B Anthropic paradox, OpenAI's AWS pivot, $600B in hyperscaler CapEx, and the EU AI Act countdown.

OpenAI breaks Azure exclusivity and lands on AWS Bedrock with managed agents, while Google starts selling TPUs off-cloud. Plus: Stanford reports agentic AI task success jumped from 20% to 77%, AWS launches its own Copilot competitor, and the enterprise cloud-AI landscape gets redrawn overnight.

Meta cuts 8,000 jobs and reorganizes around AI pods while spending over $115 billion on AI infrastructure. Today's briefing also covers OpenAI breaking Microsoft exclusivity to land on AWS, a rogue Claude agent wiping a production database, and the EU AI Act deadline holding firm.

OpenAI ends its Azure exclusivity in a landmark deal restructuring while missing revenue targets and facing IPO doubts. Today's briefing also covers Google's classified Pentagon AI deal, EU AI Act deadline chaos, and an ICLR finding that smarter AI agents hallucinate more.