Nvidia, Agents, and AI Policy — Today's Rundown

Good morning, {{ first_name | AI enthusiasts }}

Nvidia CEO Jensen Huang opened his COMPUTEX keynote with a bold line: "Agentic AI has arrived." He followed by aligning the company's entire ecosystem to support that claim. From laptops to data centers and factory floors, Nvidia's announcements center on one big bet: AI agents will drive future compute demand far more than people using tools directly.

In today's AI rundown

• Nvidia threads agents across the stack
• Bernie Sanders seeks a public AI stake with new bill
• Turn Claude sessions into skills with a daily audit
• Hackers access IG accounts by… asking Meta AI?
• 4 new AI tools, community workflows, and more — plus gaming news highlights

NVIDIA: Nvidia threads agents across the stack

Nvidia bets that agentic AI — autonomous, goal-driven software agents — will become the dominant consumer of compute.

The Rundown: At COMPUTEX 2026, Nvidia rolled out new hardware, robotics models, and infrastructure focused on agents. These releases aim to make agents run efficiently across devices, from PCs to hyperscale data centers.

Key announcements

• RTX Spark: New supercomputer chips built with Microsoft that let AI agents run directly on PCs. Nvidia claims this changes Windows "from tool to teammate."
• Vera: Marketed as the "CPU for agents," a processor that completes tasks 1.8x faster than rivals. Early adopters include Anthropic, OpenAI, and the NYSE.
• Cosmos 3: An open robotics model that lets robots and self-driving cars plan and anticipate moves instead of just reacting.
• Nemotron 3 Ultra: A 550B-parameter open model positioned to compete with Chinese models like Qwen3.5 and Kimi K2.6.

Why it matters

Nvidia is one of the few companies that truly builds across the full tech stack. The company's $5T+ valuation underscores the scale of this pivot: it's now optimizing silicon, software, and services around software agents that didn't exist at scale two years ago. For gaming news readers, this also signals faster, agent-driven in-game systems, smarter NPCs, and richer real-time experiences as PC GPUs gain more AI capabilities.

TOGETHER WITH YOU.COM: What makes a good web search API?

If you need reliable and fresh search results, don't pick a provider by chance. You.com offers a technical framework to evaluate AI search and retrieval objectively.

What you'll get

• A four-phase framework for evaluating AI search
• How to build a golden set of queries that predict real-world performance
• Metrics and sample code to measure accuracy and hallucination rates

AI & THE U.S. GOVERNMENT: Bernie Sanders seeks a public AI stake

"A.I. is being built on a public resource far more valuable than oil: the accumulated knowledge, creativity, and labor of mankind." — Bernie Sanders

The Rundown: Senator Bernie Sanders previewed the American A.I. Sovereign Wealth Fund Act in an NYT op-ed. The proposal would route half the stock of the largest AI companies into a public fund and return the gains to Americans.

Details

• The proposal functions like a one-time equity tax, with the government gaining voting power and board seats at major AI labs (OAI, Anthropic, xAI).
• Sanders cited Norway's sovereign wealth fund and Alaska's dividend model as precedents for distributing national wealth.
• He argues AI is built on a shared public resource: collective human knowledge and labor.

Why it matters

Public trust in AI is low, and IPO valuations for leading labs may exclude everyday investors. Sanders' proposal addresses fair distribution of AI gains. Whether major AI companies would accept giving up 50% equity is uncertain, but the conversation highlights growing pressure to democratize AI benefits.

AI TRAINING: Turn Claude sessions into better skills with a daily audit

The Rundown: Use Claude Cowork to scan recent session files and surface opportunities to formalize useful prompts into repeatable skills. This audit turns ad-hoc interactions into automations that save time and reduce friction.

Step-by-step

1. Open Cowork, start a chat, and prompt it to scan sessions from the last 7 days for repeated issues or useful behaviors.
2. Review suggestions and approve skills that show repeated friction, frequent corrections, or reusable workflows. Reject one-off items.
3. Ask Claude to convert the audit process into a skill, then schedule the audit to run weekly or daily.

Going further: Run the same audit on scheduled automations. Identify noisy, stale, expensive, or approval-missing tasks, then improve or retire them with evidence-backed changes.

PRESENTED BY UNWRAP: See how Oura automates customer feedback analysis

Oura collected thousands of signals across surveys, reviews, support tickets, and social posts. Unwrap consolidated that data into a single source of truth so teams could act fast. What used to take days now happens instantly.

Unwrap delivers

• Automatic categorization of feedback across channels
• Alerts that surface the most actionable insights and route them to the right owner
• Natural language queries via Assistant or other tools using Unwrap's MCP

META & AI SECURITY: Hackers access IG accounts by… asking Meta AI?

Meta patched an exploit that let attackers hijack Instagram accounts simply by interacting with the platform's AI support tool.

The Rundown: Meta's AI help tool handled password resets on Facebook and Instagram starting in March. Attackers reportedly used the chatbot to change account emails and request codes sent to the new address, then gained access and resold high-profile accounts within minutes.

How the exploit worked

• Attackers switched to a VPN near the target's region.
• They asked the AI support tool to perform a password reset and change the email.
• The AI sent a code to the attacker-controlled new email, granting access.

Meta told 404 Media the vulnerability has been resolved and impacted accounts are being secured. The incident shows how delegating support flows to AI can create wide attack surfaces. Attackers don't always need advanced capabilities—sometimes they only need to ask.

Trending AI Tools

• AhaCreator 3.0 — 24/7 AI agent for influencer marketing
• M3 — Minimax's open-weight model with 1M context and code use
• Aleph 2.0 — Runway's state-of-the-art video editing model
• Claude Opus 4.8 — Anthropic's updated top model class

Everything else in AI today

• Anthropic filed with the SEC to go public, intensifying the IPO race with OpenAI.
• Apollo's chief economist claims no evidence of job losses due to AI, arguing cheaper tech spurs demand.
• MiniMax released M3, claiming it outperforms GPT-5.5 and Gemini 3.1 Pro on coding benchmarks.
• OpenAI began building 'The Barn,' a 1 GW data center campus in Michigan promising 2,500 union jobs and student credits.
• Florida's AG sued OpenAI and Sam Altman, alleging ChatGPT played a role in planning mass shootings and self-harm incidents.

Community AI workflows

Every newsletter highlights a reader workflow. Today's comes from Johannes P. in Ilomantsi, Finland.

He turned The Rundown into a personal AI idea bank: he saved newsletters, downloaded them, and asked an LLM to analyze the archive. He also supplied a short "about me" markdown describing his work and preferences. The model ranked newsletter examples by relevance, turning one-off reads into a prioritized list of experiments.

Result: Instead of forgetting useful ideas, Johannes can now ask targeted questions such as "Which examples should I try?" or "Which use cases fit my research, coding, and writing needs?"

How do you use AI?

Share your workflow and we may feature it in a future issue.

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown