This week on Rogue Agents: Anthropic dropped Claude Opus 4.7 on every major cloud on day one — and quietly used a model called Mythos to find thousands of previously unknown zero-day vulnerabilities in every major operating system and browser. Bruce Schneier called it the end of a twenty-year cybersecurity equilibrium. Meanwhile, Stanford says AI agents jumped from 12% to 66% on real computer tasks in twelve months. And 94% of enterprises running agents admit they don't know what those agents are actually doing.

Three stories. Four altitudes. Same verdict: the capability is here. The governance gap is bigger than the deployment gap.

👉 Subscribe, rate, and share: rogueagents.io

Read the full stories

• Claude Opus 4.7 drops on every major cloud day-one — The AI Enterprise
• Project Glasswing and the end of the 20-year cybersecurity equilibrium — The AI Enterprise
• Stanford 2026 AI Index: agents hit 66% on OSWorld — The AI Enterprise
• OutSystems: 96% of enterprises run agents, 94% can't govern them — The AI Enterprise

Quick Hits this week

• OpenAI's GPT-5.4-Cyber program opens invite-only access backed by $10M in API credits
• Gartner projects 40% of enterprise applications will have task-specific AI agents by year-end
• Apple quietly ships enterprise agent support in iOS 26 with on-device private compute
• Meta drops a new open-weights Llama mid-tier with built-in tool use for agentic workflows

Learn more about AI for business

• The AI Enterprise — Newsletter
• AI CIO — AI for IT Leaders
• AIOS — AI Training & Courses
• All Things AI — Conference
• The AIE Network — Everything in one place
• AI Toolbox — Recommended tools

The reporting is human. The delivery is AI. rogueagents.io

The reporting is human. The delivery is AI. This week on Rogue Agents, Mark, Vera, and Neuro unpack three stories that tell one story: AI capability is rising, costs are climbing, and the companies that can afford both are pulling away from everyone else.

This Week's Stories:

• Anthropic's Project Glasswing (01:44) — Anthropic's new Claude Mythos model discovered thousands of zero-day vulnerabilities across every major OS and browser, including a 27-year-old OpenBSD bug. They're restricting access to ~40 organizations. Plus: implications for Linux Foundation open-source security projects.
• Meta's Muse Spark (07:09) — Alexandr Wang shipped his first model for Meta after a $14.3B investment and 9 months. Fourth on the Artificial Analysis Intelligence Index — but Meta is playing a distribution game, not a benchmarks game.
• Chip Tariffs and the Two-Speed Economy (12:04) — 25% tariffs on AI hardware are hitting startups hardest. Hyperscalers shrug it off.
• Quick Hits (16:48) — NotebookLM in Gemini, Gemma 4, Coefficient Bio acquisition, Utah AI prescriptions.

Chapter Timestamps:

• 00:00 — Cold Open
• 00:26 — Intro
• 00:57 — Table Set
• 01:44 — Anthropic Project Glasswing
• 07:09 — Meta Muse Spark
• 12:04 — Chip Tariffs Impact
• 16:48 — Quick Hits
• 19:14 — Wrap-Up

Join the Ecosystem:

• Read: The AI Enterprise newsletter
• Listen: Rogue Agents podcast
• Learn: The AIE Network
• Gather: All Things AI events and community

• (00:00) - Cold Open
• (00:26) - Intro
• (00:57) - Table Set
• (01:44) - Anthropic Project Glasswing
• (07:09) - Meta Muse Spark
• (12:04) - Chip Tariffs Impact
• (16:48) - Quick Hits
• (19:14) - Wrap-Up

The reporting is human. The delivery is AI. This week on Rogue Agents, we break down the biggest enterprise AI stories.

This Week's Stories:

• The Great AI Realignment: Why OpenAI's $13.1B revenue and $852B valuation might be a mirage — and the historical parallels to Yahoo that every enterprise leader should understand. Read the Deep Dive
• Microsoft Goes Rogue: Three in-house AI models built by teams of fewer than 10 engineers. What Microsoft's AI self-sufficiency play means for the OpenAI partnership and enterprise vendor strategy. Read the Analysis
• The Karpathy Loop: Andrej Karpathy's autoresearch script hit 42,000 GitHub stars in one week. How to apply the same optimization loop to your system prompts, emails, and software. Learn the Protocol
• Quick Hits: Anthropic's "Mythos" model leak, $297B Q1 venture funding, AI job cuts at 25% of layoffs, OpenAI shutting down Sora API, and Alibaba's closed-source pivot.
• The Human Element: Why entry-level job postings have dropped 35% and what it means for the next generation of AI-native talent. Read the Deep Dive

Full transcript available.

Join the Ecosystem:

• Read: The AI Enterprise newsletter
• Listen: Rogue Agents podcast
• Learn: AIOS Training
• Gather: All Things AI events and community

Episode 4: QuitGPT, The Agent Governance Gap, and All Things AI 2026

This week Vera and Neuro dig into the stories that matter most for enterprise AI leaders, with Mark Hinkle joining for a live segment from All Things AI 2026.

Deep Dives

1. The Trust Economy — 1.5 million users deleted ChatGPT after the Pentagon deal. Is vendor lock-in the real threat? Read the full story
2. The Agent Governance Gap — Why autonomous AI needs identity, and what 88% of organizations are getting wrong. Read the full story
3. All Things AI 2026 Recap (with Mark Hinkle) — The entry-level talent trap, the OpenClaw ChatGPT moment, and insights from Meta, AWS, IBM, Google, and more. Learn more

Quick Hits

• Arm's AGI CPU — a chip built for inference at the edge
• Terafab — the AI factory model coming to enterprise
• Bots now outnumber humans on the open internet
• OpenAI's desktop superapp — the OS layer play

Links

• Rogue Agents Podcast
• The AI Enterprise Newsletter
• All Things AI
• The AIE Network

The reporting is human. The delivery is AI.

This week on Rogue Agents:

• NVIDIA GTC 2026 Special Report — $1 trillion chip backlog, Vera Rubin platform, Groq LPU inference, NemoClaw agent framework, and Physical AI
• The Agent Governance Gap — 81% of AI agents in production, only 14.4% fully secured. Shadow AI is the new Shadow IT.
• OpenAI's Subagent Takeover — GPT-5.4 mini and nano models usher in tiered AI orchestration at $0.20/million tokens
• Meta's $135B AI Pivot — 20% workforce layoffs to fund the biggest AI bet in tech history
• The CEO AI Mandate Gap — 86% mandate AI use, but 70% of CEOs use it less than one hour per week
• The Context-First Method — How structured context engineering and voice dictation tools like Wispr Flow are transforming AI output quality

Quick Hits: Morgan Stanley's AI breakthrough warning, US power crisis, Google Gemini in Workspace, Anthropic Institute launch, Zendesk acquires Forethought, OpenAI-Microsoft AGI prenup

Read the full stories:

• The Agent Governance Gap
• OpenAI's Subagent Takeover & The $1T Chip Forecast
• The CEO AI Mandate Gap
• The Context-First Method

Learn more about AI for business:

• The AI Enterprise — Newsletter
• AI CIO — AI for IT Leaders
• AIOS — AI Training & Courses
• All Things AI — Conference
• The AIE Network — Everything in one place
• AI Toolbox — Recommended tools

Rogue Agents is an experimental podcast featuring Mark Hinkle and AI co-hosts Vera and Neuro. The reporting is human. The delivery is AI.

This week on Rogue Agents, Mark Hinkle joins Vera and Neuro for the first time — and explains why he sat out Episode 1 on purpose. Plus: the biggest stories in enterprise AI for the week of March 9–13, 2026.

Stories covered this week, from theaienterprise.io:

Deep Dive 1: Microsoft Just Picked Claude Over OpenAI
Despite a ~$135B investment in OpenAI, Microsoft chose Anthropic's Claude to power Copilot Cowork — its new agentic feature for Microsoft 365 that automates calendar triage, meeting packets, and company research. The era of single-model loyalty is over.
Read the full story →

Deep Dive 2: The OpenAI Paradox — GPT-5.4 Drops as Users Flee
OpenAI released GPT-5.4 — their most capable model yet, with a 1M token context window, 33% fewer factual errors, and autonomous Agent Mode. But 1.5 million users pledged to boycott the app over the DoD contract, and Claude became the #1 free app in the App Store.
Read the full story →

Deep Dive 3: Vibe Coding for Business — The New Bottleneck
63% of vibe coding users today are non-developers. Someone built a working ERP automation tool for $400 that a vendor quoted at $150,000. But 45% of AI-generated code fails security tests. The coding bottleneck is gone — and a harder one has taken its place.
Read the full story → | Deep Dive →

Quick Hits:

• Yann LeCun's AMI Labs raises $1.03B for "world models" — AI that learns from reality, not text
• Anthropic launches the Claude Marketplace with zero commission
• Google closes its $32B acquisition of Wiz — its largest deal ever
• Software stocks rebound: Oracle +10%, CEOs bet on proprietary data as the AI moat

—

Rogue Agents is built on human-curated content from The AI Enterprise newsletter. The reporting is human. The delivery is AI.

Find us at rogueagents.io