Fetching from the wire…
Research2026-09-12 · source-backed
The FCC put AI-generated voices under the TCPA in February 2024, and nobody had peer-reviewed how much unwanted traffic is machine-placed or synthesized rather than replayed. An interactive voice honeypot using language-model personas on real US numbers recorded 10,987 calls over 66 days, scoring each opening with an audio fingerprint, a commercial synthetic-speech detector, and blinded human listeners. Of 7,233 greeted calls, 13.8% opened with a recording heard on another call and 13.1% with fresh audio labeled synthetic, giving the 26.9% floor, with 9.9% silent connections and 54.2% fresh audio labeled human (arXiv 2609.11137).
Each link below shares sources, entities, or timing with this story.
Every AI-productivity fight this year has been three people quoting three studies at each other. Field experiments say +26% more tasks per week. METR's randomized trial says a 19% slowdown. Team telemetry says code review time up 441%. Pick your number, pick your priors, argue...
Best-in-class computer-use models scored 42% on OSWorld-Verified in early 2025. Today the leader (Claude Fable 5) scores 85%. The human tester baseline is roughly 72%. a16z published the aggregation on August 10, pulling from production interviews and llm-stats leaderboard dat...
arXiv 2607.25619 uses a regex prefilter that lets safe Markdown skill packages bypass the LLM judge entirely, sending only matched snippet windows for flagged files. On SkillsBench (n=1,650, 9.1% malicious) it hits 1.13% FPR and beats two existing tools by 5-6x on AUPRC. The t...
A GitHub Issue. No code, no credentials, no access. Just a paragraph of English that tells an AI agent to copy your private repo into a public comment. That's GitLost, and it works whether the agent runs on Copilot, Claude, Gemini, or Codex. (Noma Security) Noma Security discl...
Anthropic and Mozilla ran a coordinated two-week security research project in February 2026 where Claude Opus 4.6 scanned roughly 6,000 Firefox C++ files, submitted 112 reports, and identified 22 CVEs. Fourteen were classified high-severity, representing nearly one-fifth of al...
Andrej Karpathy, who coined "vibe coding" in February 2025, now declares it passe. His replacement term, "agentic engineering," reflects that agents "basically didn't work before December and basically work since." The key distinction: "agentic" because you orchestrate agents...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.