AI News

AI news brief: Saturday, November 7, 2026

21 stories, most important first.

OfficialResearch · · Updated

Physical Superintelligence claims an Anthropic model proved a math conjecture

Physical Superintelligence says a non-public Anthropic model proved the Aaronson-Ambainis conjecture, with the proof checked in Lean.

Why it matters: If verified, it would show a frontier model producing formal research results, but for now it is an unverified claim about an unreleased model.

2 sources
ReportedBusiness & Funding ·

Tech selloff pressures Anthropic's IPO as the company presses ahead

Coverage says a tech selloff is pressuring Anthropic's IPO and valuation, while Anthropic is still pushing forward with the listing.

Why it matters: Market conditions around AI stocks can affect the valuation and timing of Anthropic's listing, which in turn affects how investors and partners view the company.

1 source
ReportedModels ·

Meta's Muse gains Active Memory and Active Focus, and can build persistent apps

Meta says new Muse models bring Active Memory and Active Focus to Muse and Meta AI, and users can have Muse build and maintain persistent apps.

Why it matters: Persistent memory and app-building in consumer assistants show where agent products are heading. The performance claims are self-reported and untested by outsiders.

1 source
ReportedModels · · Updated

Nace AI's open Drex models return answer probabilities without generating text

Nace AI's open Drex decision models return option probabilities without generating text, and a paper reports the 27B Max model scores 67.52 on Decision Index 0.4.1.

Why it matters: Teams running classification or routing decisions can get calibrated option probabilities in one forward pass, which may cut latency and cost compared with generating answers.

2 sources
ReportedBusiness & Funding ·

Anthropic signs its first news licensing deal

THE DECODER reports that Anthropic has signed its first news licensing deal.

Why it matters: News licensing deals show that AI labs are paying for some content access, which may shape how other AI firms approach publisher agreements.

1 source
ReportedProducts ·

Anthropic blocks game cloning in Claude Code

THE DECODER reports that Anthropic has blocked game cloning in Claude Code.

Why it matters: Developers relying on Claude Code for game work may find some tasks restricted, which affects tool planning.

1 source
ReportedSafety & Security ·

OpenAI raises bug bounties and launches a security sprint

THE DECODER reports that OpenAI has raised its bug bounties and launched a security sprint.

Why it matters: Higher bounties and a focused sprint point to more active vulnerability hunting on OpenAI systems, which may matter to security teams that depend on its products.

1 source
ReportedProducts ·

Perplexity launches always-on agents for ongoing work

THE DECODER reports that Perplexity has launched always-on agents for ongoing work.

Why it matters: Persistent agents that run without a session raise the bar for monitoring and permission controls in business workflows.

1 source
ReportedResearch ·

Stanford memory system recalls long histories with 16 memory tokens

A Stanford system compresses conversation and document history into 16 memory tokens and cuts document-token use by up to 4,212x.

Why it matters: If the efficiency holds, long-running agents could carry context at a fraction of the token cost, but whether they recall rarely used details is the real test.

1 source