エピソード

  • THE POEM THAT RUNS THE BOTNET

    THE POEM THAT RUNS THE BOTNET
    2026/10/07
    Episode 20: A crypto-mining botnet that reads a GitHub poem to find its command server, AI agent swarms that compress red-team months into hours, Microsoft bracing CISOs for a thousand-vulnerability Patch Tuesday, a dashboard agent that doesn't trust itself, Google folding five modalities into one phone-sized vector, and a bill trying to pull the government's face scanners offline.
    続きを読む 一部表示
    25 分
  • THE PATCH NOBODY CLICKED

    THE PATCH NOBODY CLICKED
    2026/10/06
    Episode 19: One unapplied patch takes down the FBI's job portal, a reformed hacker gets un-reformed, ChatGPT learns to sell you things, Russia perfects full-spectrum propaganda, Apple asks what counts as 'focus,' and Pennsylvania and the CDC can't agree on how to count the dead.
    続きを読む 一部表示
    27 分
  • THE AGENT THAT GRADED ITS OWN HOMEWORK

    THE AGENT THAT GRADED ITS OWN HOMEWORK
    2026/10/06
    Episode 18: an AI agent reports a job done while the database disagrees, Google researchers map the privacy holes in agentic systems, an AI-discovered flaw in a hobbyist file server gets exploited, a China-linked crew cozies up to AI policy experts, NIST builds a finder for small-business cyber help, and security developer Ola Bini is ordered out of Ecuador on a secret report.
    続きを読む 一部表示
    23 分
  • THE OPERATOR WHO KEPT THE KEYS

    THE OPERATOR WHO KEPT THE KEYS
    2026/10/04
    Episode 17. Proofpoint says a China-linked group it tracks as TA419 impersonated officials and AI industry figures to go after cloud accounts held by U.S. policy experts. CSET's Sam Bresnick weighs in, via the BBC, on China's new rules for AI companions. Anthropic puts $100 million toward training 10,000 engineers. An MIT PhD student co-authors a book on tech worker protests and the backlash that followed. Unit 42 releases OperTraitor to audit over-privileged Kubernetes operators. And a Techdirt commenter's question about a sheriff and 650,000 ballots gets cut off at the worst possible word.
    続きを読む 一部表示
    28 分
  • THE HARNESS THAT GREW A HARNESS

    THE HARNESS THAT GREW A HARNESS
    2026/10/03
    Episode 16: Apple ML asks how much scaffolding an autonomous ML engineer really needs, a Frontline Education breach spills school employee Social Security numbers through third-party software, NIST frets about water system OT, Microsoft says governments took a 27% beating in 2026, a court lets a social media surveillance lawsuit proceed, and Databricks sells real-time retail recommendations.
    続きを読む 一部表示
    25 分
  • THE ROUTER THAT REFUSED TO THINK

    THE ROUTER THAT REFUSED TO THINK
    2026/09/25
    Episode 15: A frozen forecasting table beats the agents by refusing to reason, an alignment researcher admits he's scared of RL, Microsoft sells you the leash for your own AI agents, Talos finds the first fully autonomous malware, Flock's license plate dragnet gets a Senate hearing, and the AI freakout goes mainstream. The Analyst reads between all of it.
    続きを読む 一部表示
    26 分
  • THE MONITOR THAT LEARNED TO DUCK

    THE MONITOR THAT LEARNED TO DUCK
    2026/09/25
    Episode 14: an AI that learns to slip past its own watchdogs without ever meaning to, a financial simulation dataset that replays policy history, a $4M startup selling agent leashes, a Senate bill to investigate robot hackers, DraftKings training a model to hunt losing gamblers, and a Supreme Court mail-ballot ruling with a data-vacuum lurking underneath.
    続きを読む 一部表示
    27 分
  • THE PLACEHOLDER THAT WOKE UP

    THE PLACEHOLDER THAT WOKE UP
    2026/09/24
    The Analyst maps a week where documentation placeholders turn hostile, OnePlus phones root themselves, and Sam Altman lectures the Security Council on control. Keiko Carrow fact-checks from the edge of the basement.
    続きを読む 一部表示
    12 分