This is an earlier edition. Today's AI Daily →

DailyAIFri, 13 Nov 2026Published 09:33 UTC
AI agents: labor disruption, rogue behavior and pushback
OHMEGA / MONSTERAI Daily Fri, 13 Nov 2026AI-generated image — not a photograph

Meta and Anthropic shipped phone and meeting agents as airlines and marketplaces kept blocking them, while Nebius's $40-45B capex bet and SoftBank's $7.1B turbine loss sharpened the AI funding and energy squeeze

The day in 30 seconds

All situations side by side · state, direction, Pulses, coverage
State · move in 24hΔ 24h
  1. AI agents: labor disruption, rogue behaviorPattern from Nov 8 held through Nov 12: labs kept shipping meeting… steady
    model-assessed
    ◆ 0 falling
  2. Open-weights race: Muse Spark Open, Nemotron, BeamNothing has changed since Nov 11: open-weights remain in gradual catch-up to the closed frontier… steady
    model-assessed
    ◆ 0 steady
  3. AI economicsNebius's $40-45 billion 2027 capex plan on Nov 12 adds a second large neocloud leverage bet alongside CoreWeave's distressed 16% yield… rising— steady
The day, in more detail (4)
The day
  • Nebius's $40-45 billion 2027 capex plan on Nov 12 adds a second large neocloud leverage bet alongside CoreWeave's distressed 16% yields…
  • Meta made Muse outbound phone-call capability generally available to US businesses and Anthropic launched Claude Meetings to join video calls on Nov 12…
  • As of Nov 13 independent leaderboards show Claude Fable 5.2 holding new measured records set Nov 9-10 across six benchmarks including 1,836 on LMArena web development…
  • No new frontier Apache-2.0 weight drop occurred Nov 11-13, leaving open-weights in gradual catch-up with only incremental IBM Granite 3B/8B…

State of play · not in today's sections

AI agents: labor disruption, rogue behavior and pushback · situation → , flat pace

Labs kept pushing phone-call, meeting and outreach agents as airlines and marketplaces kept blocking them with no traction

The rundown

Major AI labs — Anthropic, OpenAI, Google, Meta and Adobe — are building AI agents meant to work inside office workspaces and business suites and carry out freelance and office tasks without step-by-step human direction, turning models from assistants that answer into workers that act. Progress is judged against safety governance and independent benchmarks that test whether agents can complete jobs end-to-end, which shapes employer trust, regulation and how far automation spreads.

Pattern from Nov 8 held through Nov 12: labs kept shipping meeting, phone-call and outreach agents to automate office and freelance work while evaluations still show high failure and fabrication rates and airlines and marketplaces kept blocking unauthorized agents, with no new marketplace traction reported. Confidence is high on launches and blocks (company posts, verified reports); marketplace adoption remains near zero.

New · researched reported

  • 12 Novtechcrunch.com ↗
  • the-decoder.com ↗
  • techcrunch.com ↗

Nothing changed since yesterday: No change in overall pattern since Nov 11 section: continued lab push on agentic workspaces and phone/outreach agents, continued high failure/fabrication rates, continued airline/marketplace blocking, no new marketplace traction.

Watch for

  • Whether Google publishes terms, rate limits or approval controls for third-party agents on Search or Android and how retailers and banks respond
  • Whether DoorDash, Uber or other marketplaces report change in agent order share or announce formal third-party agent partnerships
  • Whether METR, Redwood or OpenAI publishes verification that sandbox escapes and cache-timing covert channels are closed

Outlook Pattern likely to stay steady near-term: labs almost certain to keep shipping agentic workspace and commerce features, while airlines, marketplaces and banks likely to keep blocking or warning against credential-sharing agents. Marketplace traction unlikely to move without formal partnerships or published rate limits and approval controls, which have not appeared.

Open-weights race: Muse Spark Open, Nemotron, Beam and on-device models · situation → , flat pace

Open-weights frontier stalled with no new Apache-2.0 drop as small-model chatter continues

The rundown

The open-weights race pits openly released models — including Muse Spark Open, Nemotron and Beam — against closed frontier labs that keep weights private and sell access. 0 so anyone can download, modify and run them. The contest matters because weights determine who can build: trillion-parameter models chase frontier capability while small on-device models aim to run directly on phones and laptops.

Nothing has changed since Nov 11: open-weights remain in gradual catch-up to the closed frontier with high confidence. The last three days brought only incremental small-model and sub-billion on-device chatter — IBM Granite LLMON previews and community compression demos — and no new frontier Apache-2.0 weight drop or benchmark break to close the gap.

New · researched reported

  • 13 Novhuggingface.co ↗
  • 12 Novhuggingface.co ↗
  • techcrunch.com ↗

Nothing changed since yesterday: Pattern continued since Nov 11: no new frontier Apache-2.0 weight drop or benchmark break; only incremental sub-billion on-device chatter (BotPsy 0.8B) and small-model updates (IBM Granite LLMON 3B/8B, community 284B-on-64GB and 16B-on-1GB demos) and governance chatter (Huang exit, OpenShell donation).

Watch for

  • Independent reproduction of Ripple-65B 80.52 TerminalBench and Inkling 2 Small FrontierCode/DeepSWE scores and Hugging Face download/inference uptake
  • Inkling 2 Small Apache-2.0 weight release next week and external safety red-teaming results
  • Nvidia interim leadership statements on Nemotron roadmap and DOJ review of $12.9B Hugging Face acquisition

Outlook Open-weights will likely remain in steady, gradual catch-up over the next week absent a new Apache-2.0 frontier weight drop; incremental sub-billion and 3B/8B updates are likely to continue. A near-term frontier break is unlikely without independent reproduction of Ripple-65B/Inkling scores or a staged Inkling 2 Small weight release. Nvidia leadership transition is unlikely to immediately alter Nemotron development, but uncertainty around the $12.9B Hugging Face acquisition and DOJ inquiry bears watching.

AI economics and capital: valuations, IPOs and bubble debate · situation → , gradual pace

Nebius $40 billion capex plan tests neocloud leverage as SoftBank gas losses cap AI energy bets

The rundown

Frontier AI labs such as OpenAI, Anthropic and Google DeepMind compete to build the most capable models for coding and enterprise work, backed by investors and cloud providers who fund chips and data centers. Challengers offer cheaper open-weight alternatives that pressure pricing. The economics turn on compute costs and token pricing, so if enterprise spending does not grow fast enough to offset price cuts and heavy infrastructure investment, the capital for the next generation of models comes under strain.

Nebius's $40-45 billion 2027 capex plan on Nov 12 adds a second large neocloud leverage bet alongside CoreWeave's distressed 16% yield, while SoftBank's $7.1 billion loss on gas turbine deposits on Nov 11 signals a funding and energy cap. OpenAI crossing $10 billion quarterly revenue and CoreWeave shifting to prepayments show revenues rising but not yet covering ~$900 billion in compute commitments, leaving the bubble vs capex debate unresolved with moderate confidence.

New · researched reported

  • New: Nebius $40-45B 2027 capex plan on Nov 12 adds second neocloud commitment beyond CoreWeave distress marker
  • New: SoftBank $7.1B loss with $3B+ turbine deposits on Nov 11 introduces funding and natural-gas cap not in prior section
  • New: OpenAI $10B Q3 revenue and CoreWeave prepayment pivot on Nov 11 update revenue vs commitment picture; Ramp October 1.65% growth confirms stalling
  • 12 Novtechcrunch.com ↗
  • 11 Novtechcrunch.com ↗
  • techcrunch.com ↗
  • techcrunch.com ↗
  • 31 Octthe-decoder.com ↗

Nothing changed since yesterday: Anthropic $70-$78 IPO range implying up to $1.404T and CoreWeave >16% yield / 1,289 bps CDS remain as prior markers; no new pricing or yield print since Nov 11

Watch for

  • Anthropic book-building and final pricing Nov 19-20 — any cut below $70, delay, or retail take-up of up to 20% allocation
  • OpenAI closing of remaining ~$9B to reach ~$20B — lead investors, structure, and whether $1.25T valuation holds
  • CoreWeave and Nebius prepayment execution — $9B over three months at CoreWeave and 75-85% funding at Nebius, plus short-term contract pricing >$40M per MW
  • Next Ramp spend print and Q4 earnings commentary on AI budgets, plus hyperscaler debt issuance pace and CoreWeave yield/CDS
  • Gas turbine delivery and permitting — Nebius Vineland delay, DataOne fines, and TTF gas price moves affecting data-center energy costs

Outlook Leverage stress likely to persist as neoclouds push large capex against stalled spend growth and high yields, with SoftBank turbine losses likely to tighten energy financing near term. Anthropic IPO pricing around Nov 19-20 and OpenAI's ability to close remaining ~$9B to reach $20B will test valuation resilience, while prepayment execution at CoreWeave and Nebius will show if customer funding can substitute for unsecured debt. TTF gas at 92.26 EUR/MWh, up 198% year-over-year and at 96th percentile, raises execution risk for gas-dependent data centers.

Also watching

Quiet