AI this week: Meta’s Muse agent, a disputed math proof, and a distillation fight
A personal agent built to resist hijacking, OpenAI’s contested Navier-Stokes claim, and Anthropic’s accusations against seven Chinese labs.
September 18, 2026 · By Bohol AI
Agents moved into everyday apps, an AI math result became a controversy, and the fight over who copies whose model got public.
Meta’s Muse agent
Meta launched Muse, a personal AI agent built on Muse Spark 1.3 and controlled through the Muse app or WhatsApp. It can read and send email, browse, fill out forms and make purchases, and keeps working after you close the app. Meta designed it on the assumption that malicious web pages will try to hijack it, and built in limits so those attacks don’t take control.
A disputed proof
On September 8, OpenAI said an unreleased model, running 10,000 agents for 88 hours, produced a proof on the Navier-Stokes equations, one of the Clay Institute’s million-dollar problems. Many observers question whether the agents drew on unpublished work by two mathematicians. The result has not been settled.
Anthropic accuses seven labs
In a long report, Anthropic alleged that seven China-based AI developers, including Moonshot AI, Alibaba and DeepSeek, presented Claude’s answers to their own customers as if their models had written them, and trained their models on Claude’s output against its terms. These are Anthropic’s allegations.
Research: an assistant for the assistant
Meta researchers paired a working agent with a second agent whose only job is to remind it of earlier information it needs, which helps on long tasks.
What it means for Bohol
- Agents that send messages and make purchases for you are arriving in the chat apps people already use. Start with low-risk tasks and check what they do.
- Big AI claims can take weeks to verify. Wait for independent confirmation before repeating them.
Sources
- Meta: introducing Muse, a personal AI agent
- PBS NewsHour: Meta launches Muse
- OpenAI: Navier-Stokes
- Clay Mathematics Institute: Navier-Stokes problem
- Fortune: dispute over OpenAI’s Navier-Stokes claim
- Anthropic: threat intelligence report, September 2026
- The Batch, issue 371 (DeepLearning.AI)