Snippets.

// AI Briefing · November 14, 2026

AI Briefing

Watch on YouTube ↗Subscribe ↗

A report from Will DePue's nonprofit Unemployment Labs says six open-weight models from DeepSeek, Moonshot AI, MiniMax and Ant Group carry fingerprints showing their weights were initialized on Nvidia Blackwell GPUs, chips that US rules bar from export to China. OpenAI launched ChatGPT Wallet, which lets its "dots" agents check out at retailers on a user's behalf, starting with a limited group of US Pro subscribers aged 18 and up. In the Federal Reserve's semiannual Financial Stability Report, 87% of surveyed market contacts named AI as a salient risk to the financial system, up from 50% in the spring and 30% in fall 2025. Apache HTTP Server 2.4.71 fixes CVE-2026-119618, a critical race condition in mod_http2 that could let an unauthenticated attacker corrupt heap memory and potentially run code remotely. OpenAI broke up its standalone robotics team, moving robotics researchers under Chief Research Officer Mark Chen and the applied robotics and hardware teams into Greg Brockman's Applied organization. The FAA halted flights nationwide on November 13, its first full ground stop since January 2023, days after expanding its nearly $1 billion SMART airspace-routing AI system nationwide.

In this video

  1. 01Weight Forensics Suggest Top Chinese AI Models Were Built on Export-Banned Nvidia Blackwell Chips
  2. 02OpenAI Launches ChatGPT Wallet So Its AI Agents Can Complete Purchases for Users
  3. 03Fed's Financial Stability Report Names AI the Top Risk to the Financial System
  4. 04Apache Patches Critical HTTP/2 Flaw First Reported by Anthropic's AI Scanner
  5. 05OpenAI Disbands Its Standalone Robotics Team, Saying General Models Can Run Robots
  6. 06DHS Looks at FAA's New AI Routing Tool After a Nationwide Ground Stop

The stories

01ResearchTechCrunch ↗

Weight Forensics Suggest Top Chinese AI Models Were Built on Export-Banned Nvidia Blackwell Chips

A report from Will DePue's nonprofit Unemployment Labs says six open-weight models from DeepSeek, Moonshot AI, MiniMax and Ant Group carry fingerprints showing their weights were initialized on Nvidia Blackwell GPUs, chips that US rules bar from export to China. The method reads GPU thread layouts that leave a pattern in published weights. The report calls the evidence conclusive for DeepSeek V4-Flash and "strong" for the others, and the authors say they reproduced the patterns on real B300 and GB200 hardware. It cannot show where the chips were located or whether any export law was broken, but it puts new pressure on Nvidia and on export-control enforcement during the company's leadership crisis.

02ProductTechCrunch ↗

OpenAI Launches ChatGPT Wallet So Its AI Agents Can Complete Purchases for Users

OpenAI launched ChatGPT Wallet, which lets its "dots" agents check out at retailers on a user's behalf, starting with a limited group of US Pro subscribers aged 18 and up. Cards are saved once and charged through payment tokens, so neither the agent nor the merchant sees the card number, and each purchase needs a biometric approval. OpenAI plans spending limits and pre-approved purchase categories that would let agents buy more autonomously. It is one of the most direct moves yet toward agents spending real money, though some retailers, Amazon among them, block AI agents.

03PolicyCNBC ↗

Fed's Financial Stability Report Names AI the Top Risk to the Financial System

In the Federal Reserve's semiannual Financial Stability Report, 87% of surveyed market contacts named AI as a salient risk to the financial system, up from 50% in the spring and 30% in fall 2025. Respondents pointed to stretched equity valuations, cyber risk from state actors or frontier models themselves, and complex AI-related financing. The Fed compared the current AI infrastructure debt and investment cycle to the late-1990s telecom broadband boom. When the central bank flags AI as the top concern, it adds official weight to worries about an AI bubble.

04IndustryBleepingComputer ↗

Apache Patches Critical HTTP/2 Flaw First Reported by Anthropic's AI Scanner

Apache HTTP Server 2.4.71 fixes CVE-2026-119618, a critical race condition in mod_http2 that could let an unauthenticated attacker corrupt heap memory and potentially run code remotely. Anthropic's AI-based OSS Scanner reported it first, and Google Cloud researchers reported it independently days later. Only servers with HTTP/2 enabled are exposed, and it is off by default. Wiz telemetry found vulnerable versions in 64% of cloud environments, and the case is another example of AI tools finding serious bugs in core internet infrastructure.

05IndustryFortune ↗

OpenAI Disbands Its Standalone Robotics Team, Saying General Models Can Run Robots

OpenAI broke up its standalone robotics team, moving robotics researchers under Chief Research Officer Mark Chen and the applied robotics and hardware teams into Greg Brockman's Applied organization. OpenAI says robotics remains core and that general-purpose models such as GPT-6 Astra can be applied directly to robots. The move is a bet that robotics is a downstream application of general intelligence rather than a specialized problem, a view some robotics researchers dispute. OpenAI is still developing its own humanoid and is targeting a public demo in 2027.

06PolicyFuturism ↗

DHS Looks at FAA's New AI Routing Tool After a Nationwide Ground Stop

The FAA halted flights nationwide on November 13, its first full ground stop since January 2023, days after expanding its nearly $1 billion SMART airspace-routing AI system nationwide. Transportation Secretary Sean Duffy blamed "a communication issue," and DHS head Markwayne Mullin said of SMART, "We're looking at that." Contractor Air Space Intelligence says the ground stop was not caused by SMART. No link has been shown, but the episode raises questions about putting AI tools into safety-critical public infrastructure.

Previous digests

Let's talk on WhatsApp