Skip to content

Vol. I · No. 20Friday, October 23, 2026dailyprompt.si

The Daily Prompt

Superintelligence, delivered daily

No. 19 ·

Two critical PHP flaws, found with AI help, send websites and governments scrambling

AI models helped researchers find decade-old bugs in the language behind most websites, while Anthropic opened a waitlist for a proactive agent and Meta said 10 million people now use its Muse agent.

Top news

  1. AI helped find two critical PHP flaws; sites go dark while admins patch

    PHP's maintainers released fixes on Oct. 21 for two flaws that can let an unauthenticated attacker run code on servers running PHP, the language behind roughly three-quarters of websites whose language can be identified. Wiz and Calif used OpenAI's GPT-5.6 Sol and Google's Gemini 4 Argon to find and verify the bugs, which date back to PHP 5. Finland, the Netherlands, France and Denmark pulled some government services offline as a precaution.

    Source: Computer Weekly; PHP.net; OpenAI

  2. Anthropic opens a waitlist for Claude Duet, a proactive agent

    Claude Duet is a research preview in which Claude surfaces things that need your attention, takes on tasks and keeps working when you leave. It is starting with U.S. Pro and Max subscribers by waitlist, on claude.ai and as an iOS and Android app. Anthropic says it uses the same usage limits as the rest of Claude.

    Source: Anthropic

  3. Meta says 10 million people use its Muse agent, adds new stock to fund AI

    On its third-quarter earnings call, Meta said more than 10 million people in the U.S. and Canada set up Muse personal agents in about six weeks. Revenue was $64 billion, up 26%, while costs rose 68%. Meta also plans a new class of non-voting shares, issued as a stock dividend, that it can use to raise money, pay staff and buy companies.

    Source: TechCrunch

  4. Google DeepMind engineer quits over military AI contracts

    Research engineer Sofia Liguori resigned from Google DeepMind in protest of Google's military contracts with the U.S. and Israel, The Intercept reported. She said internal channels for raising concerns did not work. Google said public-sector contracts go through a review process and that employees have dedicated channels to raise concerns.

    Source: The Intercept

  5. Amodei defends Anthropic's 'race to the top' and admits some missteps

    Responding to Encode AI's Nathan Calvin, Anthropic CEO Dario Amodei listed nine decisions he said cost the company commercially but made AI safer, including delaying broad release of its Mythos-class models by four months. He conceded three points could have been handled better and said more on the 'pacing' of AI development is coming. The post lands as Anthropic heads toward a reported IPO.

    Source: The Next Web

New tools in this edition

  • Recursive FlintCoding

    A strong option for private, offline coding, but the full weights are about 80 GB and you have to set up and run the model yourself.

  • ElevenLabs Eleven RealtimeVoice

    A leading choice for natural-sounding voices, but credits run out fast on long projects and higher tiers get expensive.

  • UseIt StudioDesign

    Covers the step most AI image tools skip, getting a design ready for production, but it is early software and its pricing is not published.