Thursday, 22 October 2026
Live
ZOTPAPER

News without the noise

ZOTPAPER

Zotpaper: today's edition

Security

Millions of exploit attempts after critical PHP flaws disclosed, Cloudflare says

FBI warns of remote code execution risk; multiple organizations take websites offline as a precaution

Cloudflare CEO Matthew Prince warned Thursday that the company's network detected millions of exploit attempts following the public disclosure of two critical vulnerabilities in the PHP programming language, with thousands of attempts observed even before the vulnerabilities were publicly announced.

Zotpaper··2 min·2 sources
SignalBetaWhat the people building AI are saying
Today's Signal
  1. 7 voices
    AI-Discovered PHP flaws Trigger Global website Shutdown as Governments Race to patch
    Cloudflare has deployed default web application firewall (WAF) protections for newly disclosed PHP vulnerabilities before public disclosure. The rules use request and response signals to stop exploit attempts and prevent exposed data from reaching attackers.
    Cloudflare, · On X
  2. 2 voices
    OpenAI to bring AI models to Google Cloud, expanding beyond Microsoft
    OpenAI models are coming to Google Cloud. We're bringing our frontier models and Codex agentic workflows to Google Cloud's Model Garden, so enterprises can use our technology where they already build.
    OpenAI, · On X
  3. 2 voices
    SpaceX AI revenue surpasses Starlink as earnings beat estimates, capex set to double
    Q3 2026 highlights: - Delivered revenue of $14.6 billion, up 161% year-over-year - Successfully launched our first revenue-generating Starship mission, delivering 26 Starlink V3 satellites to our broadband constellation, representing approximately 10x the downlink capacity added by a Falcon 9 Starlink launch - Added $6.7 billion of incremental Super Intelligence solutions and infrastructure revenue year-over-year, driven by cloud services agreements, the acquisition of Cursor, and our growing SI models and software applications, including @Grok @Bot - Delivered year-over-year revenue growth of 69% and income from operations growth of 99% for the Connectivity segment, supported by strong momentum in enterprise and government adoption, and continued @Starlink consumer subscriber growth - Continued to scale nameplate compute, reaching 1.7 GW as of September 30, 2026
    SpaceX, · On X

Tech

AI

Business

Research Threads

All threads →

Latest Research

All research →
Weekly Digest · 70 papers ·

This Week in AI Research: Better Signals Beat Bigger Models

Across agents, evaluation, robotics and safety, progress came from supervising the right intermediate state rather than scaling blunt end metrics.

Threads:Agent Security & Attacks 136Generative Video Compositing 4Post-Training Quantization 7Agent Harness Optimization 176Low-Precision Training 2All threads →
Paper · Academic

System streams infinite 3D worlds that remember and restore user edits

Xiaoxiao Ma et al. · University of Science and Technology of China, Joy Future Academy, JD

The authors present WorldStream, an open-ended 3D world model that concurrently extends terrain beyond any pre-specified map, acts on open-ended language in the running world, and restores visited places with their edits. It uses a hierarchical planner (Macro/Micro) over a typed decision model and deterministic local code to generate terrain, roads, settlements, and entities, all coordinated through a single persistent world record. The key finding is that streaming keeps per-frame cost bounded as explored extent grows, while the retained record ensures consistent state across planning, generation, interaction, and re-rendering.

Paper · Industry

Persistent memory in LLMs enables cross-session security breaches

Pedro H. Barcha Correia et al. · Universidade de São Paulo (USP), Universidade do Estado de Santa Catarina (UDESC)

The authors propose a general threat model for memory injection attacks on large language models (LLMs) that leverage persistent memory features. They show that by planting malicious instructions in personal memories, projects, skills, organization instructions, or Model Context Protocol (MCP) resources, attackers can achieve all NIST-defined attack goals, availability, integrity, privacy, and misuse, across ChatGPT Instant 5.3, Claude Sonnet 4.6, and Gemini 3 Flash. The attacks persist across sessions and, in the case of Claude, can spread to other users in an organization.