---
title: 'In the News: November 4, 2026 (Midday)'
description: 'Conductor Cloud goes read-only November 13 and shuts down November 20 after Conductor announced it is joining Meta. Export or transfer workspaces first.'
canonical_url: 'https://darkfactory.dev/news/2026-11-04-midday'
markdown_url: 'https://darkfactory.dev/news/2026-11-04-midday.md'
collection: news
date_published: '2026-11-04T16:20:00-05:00'
date_modified: '2026-11-04T16:20:00-05:00'
---

# In the News: November 4, 2026 (Midday)


Conductor, the app for running teams of coding agents, is closing its cloud product within three weeks of announcing that it is joining Meta. Anthropic also released a provable-inference prototype, and its write-up explains how a human can audit work the company credits to Claude.

## 1. Conductor says it is joining Meta and will shut down Conductor Cloud on November 20

**[Conductor is joining Meta](https://www.conductor.build/announcement)** · Charlie, Conductor · Conductor announcement, November 4, 2026

The post, signed only "Charlie", gives a dated timeline. As of November 4, new users cannot sign up or download the app, free users cannot upgrade, and pro users keep pro features for free. On November 13, Conductor Cloud goes read-only and active subscribers are refunded. On November 20, deletion of Cloud user data begins. Local Conductor keeps receiving new models and critical bug fixes until early 2027, then winds down. The post offers a "Transfer to local" action for cloud workspaces and an export under Settings, Account, Export data. It says "We have never sold or trained on Conductor user data", and promises something new "early next year" without detail. The page gives no terms for the Meta deal. On the Hacker News thread (30 points, 16 comments at about 4 PM EST), one commenter called the cloud timeline "very aggressive".

**Why it matters:** If your agent work lives in Conductor Cloud workspaces, the dates that count are November 13 and November 20. Move the workspaces to local or export them before then. Keep the specs and scripts that drive your agents in your own repository, so a tool shutdown costs you an interface and not your process.

## 2. Anthropic releases a provable-inference prototype and rests its trust on Lean statements a human can audit

**[Progress on provable inference for large language models](https://www.anthropic.com/research/provable-inference)** · Roy Rinberg, Alex Allain (Anthropic) and co-authors · Anthropic research, November 2, 2026

The post describes FlashProve, a system that attaches a cryptographic proof to a model's response showing a specific model produced it. For Llama 4 Maverick (400B parameters, 4-bit weights), it reports about 110 tokens per second on one H100 and a proof of roughly 19 MB per 1,024-token request. That is about 70 to 95 times the GPU time of running the model on vLLM. The authors write that "Claude is good at optimization, but often makes decisions that are hard to evaluate", and respond by stating the security guarantees as top-level Lean statements. Code, Lean proofs and a spec are released under Apache 2.0, with a fast C++ verifier and a slower Lean one. The post lists its limits: a research prototype, contexts up to 1,024 tokens, proofs about an integer version of the model under greedy decoding, and no guarantee that the fast code matches the Lean protocol.

**Why it matters:** The stated method for agent-built code that is hard to review is to shrink what a person must check to a few formal statements and keep a slow, verifiable version beside the fast one. The authors say those statements still need a manual audit and the fast code is not covered, so the review gets smaller, not absent.

---

## Also this cycle

- **[Introduction to Harness Engineering for Vulnerability Research](https://grahamhelton.github.io/harness-engineering-workshop-book/introduction.html)** · Graham Helton and Ryan Basden · Workshop book, November 4, 2026 · A free workshop in which participants save a baseline agent response, build a harness around an authorized open-source target and test one hypothesis. The introduction states "Agent output is a lead, not a finding." It is a teaching method, and the pages read report no measurements.