Skip to content
Source · Daily Brief

AI Daily Brief — 28 September 2025

Sunday was a transit day between the closing week and the loaded Monday ahead. Claude Sonnet 4.5 sat on Vertex AI listings ready to drop. DeepSeek V3.2-Exp was pre-positioned on Hugging Face and GitHub, with its Sparse Attention efficiency story already leaking. California SB 53 awaited Governor Newsom’s Monday signing. OpenAI’s ChatGPT Pulse kept generating weekend coverage as the first major shift from reactive chatbot to anticipatory daily briefings. Stanford and BetterUp’s “workslop” study went viral — finding 41% of workers had received AI-generated content that looks polished but lacks substance, costing ~2 hours of rework per instance and roughly $9M/year for a 10,000-person org.

Top stories

  • “Workslop” study goes viral over the weekend. HBR-published BetterUp Labs + Stanford research coined the term: AI-generated content polished but empty, 41% of workers received it, ~2 hours rework per instance, ~$9M/year cost for a 10K-person org. Spread fast through tech Twitter/LinkedIn over Sep 27-28. via Harvard Business Review
  • Claude Sonnet 4.5 pre-positioned for Monday. Vendor leaks and Vertex AI listings pointed to an imminent Anthropic launch — “best model in the world for real-world agents, coding and computer use” at unchanged $3/$15 per Mtoken with up to 64K output. via Anthropic Help Center
  • DeepSeek V3.2-Exp staged for Monday. Open-weight successor to V3.1-Terminus introducing DeepSeek Sparse Attention (DSA) via lightning indexer + fine-grained token selection — promising 50%+ long-context API cost cuts and 70-80% cache-hit savings. via TechCrunch
  • California SB 53 signing imminent. Governor Newsom’s expected Monday signing — first enforceable US frontier-AI safety law, targets models trained above 10^26 FLOPs (OpenAI, Anthropic, Google DeepMind, Meta), with $1M-per-violation penalties from Jan 2026. via Office of Governor Newsom
  • Bartz preliminary approval continues to ripple. Weekend coverage continued unpacking Anthropic’s $1.5B settlement — the first-ever class settlement in the generative-AI copyright lawsuit wave. via Launch Consulting

Who shipped

Nobody Sunday. Anthropic, DeepSeek and the California governor all queued for Monday. OpenAI, Google DeepMind, Meta, xAI stayed dark.

Open-source pulse

No new releases. DeepSeek V3.2-Exp staged on Hugging Face but not yet announced. Alibaba’s Qwen3-Max + Qwen3-Omni from the prior week continued to draw weekend coverage as China’s two top frontier-scale releases.

Money, infra & hardware

Sunday analyses continued unpacking NVIDIA’s Sep 18 deal with Intel to co-develop custom data-center and PC silicon over NVLink, plus the Sep 19 $900M+ Enfabrica acqui-hire (CEO + staff hired, core interconnect tech licensed) — moves reshaping the AI-fabric stack as Stargate progresses.

Quiet corners

CAMIA (Context-Aware Membership Inference Attack) — a new privacy attack that determines whether a specific person’s data was used in LLM training — circulated in weekend AI-news roundups as a sharper-than-prior membership-inference baseline. No EU AI Act movement. No US Congress activity.

By the numbers

  • 0 — fresh launches dated Sep 28
  • 41% — workers receiving “workslop” per HBR/Stanford
  • ~$9M/year — workslop cost estimate for a 10K-person org
  • 3 — major Monday landings queued (Sonnet 4.5, V3.2-Exp, SB 53)
  • Most-mentioned topic in weekend coverage: workslop
  • Quietest segment: fresh model releases

Compiled by AI Feed’s editor from verified web sources for 28 September 2025.