Signal SentrySearchUpdated Nov 13, 17:30 UTCPM Drop
Drop #084 · Fri, Nov 13, 2026 · Evening

ChatGPT agents can now pay; LLM judges miss bad writes

10-second version
  1. OpenAI's ChatGPT Wallet lets its dots agents buy for some US Pro users, who approve each purchase.
  2. A preprint finds an LLM judge passed agents' forbidden writes that a simple rule-based check caught.
  3. Brave now sells search to agents per call over x402, and llama.cpp reaches a stable v1.0.0.
Our read

What changed

Every signal →

Our reading of this Drop's sources: what to rethink, and one thing worth checking this week. Not a prediction.

  1. Agents are getting their own ways to pay

    If you assumed
    that an agent needs a person's account or API key before it can buy anything
    Now test
    which of your paid APIs or checkout pages an agent could pay for directly, and who approves each charge
    This week
    List where your agents could spend money, the cap on each, and who signs off before a charge goes through

    Based on:OpenAI's ChatGPT Wallet lets its dots agents buy for some US Pro usersBrave Search API takes per-request x402 payments from agents, no account needed2 primary sources

  2. An LLM judge is not a safe last check on agent writes

    If you assumed
    that an LLM judge reading the transcript will catch an agent's forbidden actions
    Now test
    whether a plain rule check on expected versus actual writes flags failures your judge passes, if the preprint's finding holds for you
    This week
    Log every database write your agents make and compare it with the writes each task should produce

    Based on:Preprint: LLM judges passed forbidden agent writes that a simple rule check caught1 primary source

The Drop in 60

This Drop in 52 seconds

Tap to play. It never plays by itself, and the captions are in the picture.

Transcript

OpenAI launched ChatGPT Wallet. Its dots agents can buy things for some US Pro users, who approve each purchase. A preprint finds an LLM judge passed forbidden agent writes. A simple rule check caught all 11. Brave says agents can now buy its search per call over x402, no account needed. OpenAI's Daybreak agents test web apps and draft patches, and engineers review them, SecurityWeek reports. And llama.cpp shipped its v1 release; future API breaks will come with a major version bump. Agents are starting to spend money. This week, list where yours could spend, and who approves each charge.

Made with AI from this Drop's sourced items. The facts come from the sources linked on each item. Script: Claude · Voice: OpenAI gpt-4o-mini-tts · Visuals: motion graphics

442models priced on OpenRouter
+213★stars today on the top repo
4,162open roles at AI labs
15 ddays to the next AI-law deadline
Ranked by impact

Top 5 that matter

1
Agents

OpenAI's ChatGPT Wallet lets its dots agents buy for some US Pro users

OpenAI built ChatGPT Wallet with Amex, Mastercard and Visa, CNBC reports. Dots pay with a scoped, per-purchase payment token rather than card details, and users approve each buy with a check such as Face ID, TechCrunch says.

Why it matters: Agent checkout is live for real shoppers, with a person approving each purchase. If you sell online, check whether your site blocks OpenAI's agents; OpenAI published merchant guidance.

2
Agents

Preprint: LLM judges passed forbidden agent writes that a simple rule check caught

Scoring 2,160 tau2-bench episodes four ways, the author reports an LLM judge agreed with native scoring 56.2% of the time and passed 10 to 11 of 11 forbidden writes even with the policy; a rule-based boundary check caught all 11.

Why it matters: If an LLM judge is your only check on a live agent, it may pass the writes that matter most. Test a plain rule: was a write expected, and did one happen?

3
Agents

Brave Search API takes per-request x402 payments from agents, no account needed

Brave says agents can pay per Search API call in USDC on Base via x402: an unpaid request gets an HTTP 402 Payment Required reply with the price, the agent signs a payment and retries. Each search costs $0.005.

Why it matters: Pay-per-call removes API key management but puts money in the agent's hands. If you try it, decide who funds the agent's wallet and what caps it.

On the Frontier watch →
4
Security

OpenAI's Daybreak uses agents to pen test apps and draft patches for engineers

SecurityWeek reports OpenAI's Daybreak Red Team uses agents to probe web apps in isolated environments, and Defense Factory takes findings from it or Wiz, Snyk and Endor Labs, drafts patches and opens draft PRs for review.

Why it matters: Security vendors now ship agents that write fixes, yet OpenAI keeps merges with people. If you pilot it, check which repos and endpoints the agents may reach.

5
Models

llama.cpp reaches v1.0.0 and ties any API break to a major-version bump

The llama.cpp team at Hugging Face released the llama.cpp and ggml libraries as v1.0.0 and says API or ABI breaks will now come with major-version bumps. It supports more than 150 architecture variants, the team says.

Why it matters: Apps built on local inference can pin a major version and expect stable interfaces within it. Check how your builds pull llama.cpp before moving across majors.

Models

New in models

See all Models →
Models

Edge0 open-sources a 145B model it says runs on a 48GB laptop

Why it matters: If these numbers hold, some agent coding could move off cloud tokens. They are the vendor's internal results, so test it on your own tasks before relying on it.

More detail: Edge0 open-sources a 145B model it says runs on a 48GB laptop

Edge0 says about 10B of Edge0-145B's parameters are active per step, and its laptop test hit 80+ tokens/s at 28GB peak memory. Released with Edge0-120B, it scores 64.6 on DeepSWE 1.1 in Edge0's own evaluations.

Agents

Agents at work

See all Agents →
Agents

Databricks routes Codex through Unity Gateway so developers never hold the OpenAI key

Why it matters: Company-wide coding agent rollouts are becoming a gateway question. Check whether your developers hold raw model keys and whether spend and requests are logged per team.

More detail: Databricks routes Codex through Unity Gateway so developers never hold the OpenAI key

In a post on its OpenAI partnership, Databricks says Codex can sign in with a Databricks credential while the OpenAI key stays inside Unity Gateway, which logs each request and lets leaders set budgets by team.

Agents

Block mixes models across agents in its Buzz workspace, Databricks says

Why it matters: The question is moving from best model to best mix. Block shares no results, so treat it as a pattern: benchmark a coordinator with smaller workers against one model.

More detail: Block mixes models across agents in its Buzz workspace, Databricks says

Databricks says Block built Buzz, an open-source, self-hostable space where staff and agents work in shared channels, with Unity Gateway giving model choice per agent. Block is testing a strong model coordinating smaller ones.

Agents

Preprint: delegating to sub-agents gave no consistent accuracy gain in generative search

Why it matters: Adding sub-agents is not a free upgrade. Before splitting a search task across agents, compare it with one model calling the tools directly on questions past its cutoff.

More detail: Preprint: delegating to sub-agents gave no consistent accuracy gain in generative search

Researchers report that delegating to tools and other LLMs gave no consistent accuracy gain over non-delegated setups, and could cut accuracy sharply on post-cutoff questions, where the same LLM using tools directly did much better.

On the Frontier watch →
Business

Money and deals

See all Business →
Business

LangChain raises $300M at a $3B valuation for its agent platform

Why it matters: Investors are backing agent runtime, evals and governance, not just frameworks. If you build on LangChain, watch the November 20 release for LangSmith changes.

More detail: LangChain raises $300M at a $3B valuation for its agent platform

LangChain says CapitalG led the round and over 7,000 organizations use LangSmith, its agent platform; it says Workday now ships a new agent in under a day, down from five months. More LangSmith features are due November 20.

Business

Huang's exit puts Nvidia's open-model support and Hugging Face deal in question

Why it matters: If you rely on Nemotron or Hugging Face, nothing has changed yet, per TechCrunch. Watch for word from Nvidia's interim leadership on the deal before planning around it.

More detail: Huang's exit puts Nvidia's open-model support and Hugging Face deal in question

TechCrunch reports Nvidia has yet to finalize its $13 billion Hugging Face acquisition, and it is unclear how interim CEO Colette Kress will treat it. An insider told TechCrunch they expect no change to Nemotron work.

Laws

Law watch

See all Laws →
LawsFollowing since Nov 12, 2026in 4 Drops

Lutnick agrees to testify on Nvidia's China chip sales; Senate asks Huang for Dec. 3

Why it matters: Scrutiny now covers how Commerce granted export licenses, not just Nvidia. If your plans depend on chip supply in Asia, track the December hearings for licensing changes.

More detail: Lutnick agrees to testify on Nvidia's China chip sales; Senate asks Huang for Dec. 3

Commerce Secretary Lutnick agreed to testify to the Senate Banking Committee on Dec. 10, and Chair Tim Scott asked Huang to appear Dec. 3, CNBC reports, citing NBC. Democrats Wyden and Neal opened their own inquiry, TechCrunch says.

Today's Talk

What people are arguing about

See all Today's Talk →
Today's Talk

HN threads track a Claude and Claude Code outage; top thread at 398 pts

Why it matters: An outage at one provider can stall every agent built on it. Check whether your coding and agent workflows have a fallback model, or a plan for a day without one.

More detail: HN threads track a Claude and Claude Code outage; top thread at 398 pts

A Hacker News thread on elevated errors at claude.ai, Claude Code and platform.claude.com drew 398 pts and 333 comments; a Claude Is Down thread drew 178 pts, and another flagged Claude OAuth failures.

Play the NewsAI ConnectionsAbout 3 minutes, built from this Drop's items.