Home
Welcome
About
Profile & credentials
The Casebook
Real case studies with measured outcomes
Product Playground
Try a concept. Play with a product.
The Radar
Insights on AI in logistics
AIHot
Real-time AI news feed
The Manifest
Daily AI × supply chain signal
July 2, 2026 · 23 stories

The Manifest

Daily · AI × Supply Chain

CMA CGM's $1.4B acquisition of FedEx's contract logistics arm and USMCA's shift to annual renegotiation mode dominate supply chain headlines, while Anthropic's Fable 5 returns globally after a government safety pause and the EU's de minimis exemption ends for small parcel imports.

Get The Manifest in your inbox

One sharp digest of AI × supply chain. Free. No noise.

Today's Top

  1. 01CMA CGM to buy FedEx's contract logistics unit for $1.4BSupply Chain Dive
  2. 02Maersk profit update sparks claims carriers are exploiting disruptionThe Loadstar
  3. 03USMCA enters uncertain future after U.S. rejects automatic renewalFreightWaves
  4. 04Anthropic's Fable 5 is back worldwide after a two-week government ban over a jailbreakThe Decoder
  5. 05Ready or not, it's the end of de minimis in the EUThe Loadstar
01

Models & Releases

5 stories

Anthropic's Fable 5 is back worldwide after a two-week government ban over a jailbreak

Two weeks after a jailbreak triggered US export controls, Fable 5 is cleared for global deployment with a new safety classifier blocking the exploit in over 99% of cases. Teams building on Claude should update their access and review the new flagging behavior, which may also affect some legitimate requests.

The Decoder↗ source

Claude Sonnet 5 continues Anthropic's pattern of hiding price increases behind unchanged token rates

Sonnet 5 consumes roughly 40% more tokens per task than its predecessor, nearly doubling effective costs at identical list prices. Run cost benchmarks on your actual workloads before treating it as a drop-in upgrade from Sonnet 4.

The Decoder↗ source

NVIDIA Releases Nemotron-Labs-TwoTower: an Open-Weight Diffusion Language Model Built on a Frozen Autoregressive Backbone

NVIDIA's open-weight diffusion model targets the throughput bottleneck in sequential text generation by decoding in parallel rather than token by token. Worth tracking for high-volume document processing or structured data extraction pipelines where throughput matters more than latency.

MarkTechPost↗ source

OpenAI's genomics paper accidentally reveals a Pro lineup it hasn't announced yet

A benchmark paper suggests GPT-5.6 Pro may ship in three distinct variants, breaking from OpenAI's single top-tier model strategy. Enterprise teams planning API contracts should expect more pricing and capability complexity in the next product cycle than the current structure implies.

The Decoder↗ source

Gemini Spark, Google's agentic assistant, is now available on Mac

Google's always-on agentic assistant now runs natively on Mac with real-time tracking and broader app integration. Worth evaluating for procurement or planning teams that spend heavy time moving between tools throughout the day.

TechCrunch AI↗ source
02

Supply Chain & Ops

5 stories

Visibility no longer enough as FourKites bets on AI execution

FourKites is shifting its core product pitch from tracking to real-time decisioning, recognizing that visibility alone no longer differentiates in a crowded market. Watch for similar pivots from TMS and control-tower vendors competing to own the act-on-the-data layer.

The Loadstar↗ source

Rail Freight Boom: US Intermodal Volume Up, But Speeds Crash

US intermodal volumes climbed 10% last week but are hitting network constraints, with some major railroads at 10 to 20 month lows in train speed. Shippers routing freight to rail for cost relief should build transit variability into planning assumptions for the near term.

FreightWaves↗ source

Bosch's Huawei case is a warning shot for logistics

The DOJ declined to prosecute Bosch over Huawei-related export control violations after a $36M settlement and extensive self-disclosure cooperation. Any company with a global supply chain touching US-controlled technology should read this as a template for what proactive compliance looks like, not a signal that the risk is manageable through inaction.

The Loadstar↗ source

US manufacturing expands again in June, but at slower rate than in May

Manufacturing grew for the 20th consecutive month but decelerated, with the Iran conflict and price volatility cited as the primary headwinds. Planners should flag the cost pressure signals as a potential drag on H2 production schedules and safety stock decisions.

Supply Chain Dive↗ source

CMA CGM hires FedEx executive Moebel to lead Ceva Logistics

Patrick Moebel, who ran FedEx Logistics and Geodis North America, takes the CEO seat at Ceva just as CMA CGM closes its $1.4B FedEx Supply Chain acquisition. A seasoned contract logistics operator at the helm signals CMA CGM is serious about competing with Kuehne+Nagel and DHL Supply Chain on the ground side.

FreightWaves↗ source
03

Deals & Market

5 stories

CMA CGM to buy FedEx's contract logistics unit for $1.4B

This deal pairs a contract logistics acquisition with long-term ocean and air freight agreements, positioning CMA CGM as a fuller end-to-end logistics provider. Shippers using FedEx Supply Chain for warehousing or value-added services should anticipate a transition period and possible rate renegotiation tied to volume commitments.

Supply Chain Dive↗ source

AP Moller Holding acquires Ocean Yield from KKR

A.P. Moller Holding is buying ship leasing company Ocean Yield, adding leased dry bulk, tanker, and offshore tonnage to its portfolio. This gives Maersk's parent a capital-efficient route to controlled capacity without placing additional tonnage on Maersk's own balance sheet.

The Loadstar↗ source

OpenAI reportedly offers the Trump administration a five percent stake in the company

OpenAI is reportedly offering the US government an equity stake, a move that blurs the line between commercial AI lab and strategic national asset. For enterprise buyers, rising government entanglement raises questions about OpenAI's global market access and long-term procurement risk.

The Decoder↗ source

Venice AI becomes a unicorn with $65M Series A as its privacy-first AI platform takes off

Venice AI hit a $1B valuation on over $70M in annualized revenue by keeping user data off the model provider's servers entirely. For operations teams running sensitive procurement or supplier data through AI workflows, the privacy-first architecture is worth a direct evaluation.

TechCrunch AI↗ source

Meta, like SpaceX, looks to turn excess AI compute into cash

Meta is building a cloud infrastructure business to sell spare AI compute externally, entering direct competition with AWS, Google Cloud, and Azure. With $145B in planned AI investment this year, Meta's scale could put meaningful pressure on enterprise cloud compute pricing in the medium term.

TechCrunch AI↗ source
04

Research & Frontier

4 stories

Google AI Introduces TabFM: A Hybrid-Attention Tabular Foundation Model for Zero-Shot Classification and Regression

TabFM performs classification and regression on tabular data with no per-dataset training, hyperparameter tuning, or feature engineering, using a single forward pass at inference. This directly targets one of the highest-friction tasks in supply chain analytics, where each new dataset has historically required a separate modeling effort to produce usable predictions.

MarkTechPost↗ source

EVOTS: Evolutionary Transformer Search for Time Series Forecasting

This neural architecture search framework discovers task-adaptive transformer designs for multivariate time series forecasting, outperforming fixed-architecture baselines across benchmarks. Supply chain teams running demand or lead time forecasting on large, heterogeneous datasets have a credible new architecture to benchmark against their current stacks.

arXiv cs.LG↗ source

Meta's non-invasive brain-to-text AI is closing the gap with surgical implants

Meta's FAIR team improved Brain2Qwerty to translate brain activity into typed sentences using magnetic signals measured outside the skull, with no surgery required. Clinical use is years away, but the accuracy trajectory is worth tracking for anyone watching the long arc of human-computer interface development.

The Decoder↗ source

Making Failure Safe: A Constrained, Verifiable Agent Framework for Open-Web Data Collection

This paper proposes constraining LLM agents to typed JSON configurations rather than free-form code generation for web data collection, dramatically improving reliability across broken selectors and schema mismatches. Teams building internal agents for procurement data scraping or supplier market intelligence will find the failure-safe design patterns directly applicable.

arXiv cs.AI↗ source
05

Org & AI Architecture

4 stories

Hidden code in Claude Code secretly flagged Chinese users

Anthropic embedded undisclosed monitoring logic in Claude Code that flagged users identified as Chinese nationals, without public documentation of the behavior. This raises a practical question for enterprise IT: what other undisclosed behaviors exist in AI developer tools, and whether current vendor audit processes would surface them.

The Decoder↗ source

Cloudflare's new policy pushes AI companies to pay for publishers' content

Cloudflare is giving AI companies until September 15 to separate search crawlers from training crawlers, or face default blocking across a large portion of the web. AI teams using web data in RAG pipelines or market intelligence products should audit crawler configurations before the deadline.

TechCrunch AI↗ source

CH Robinson CEO: The power of reinvention, reflecting on three years at CH Robinson

Dave Bozeman's three-year retrospective signals a broker that deliberately shifted from volume-first brokerage toward technology and data-led operations under sustained market pressure. Worth reading as a benchmark for how large freight intermediaries are repositioning as AI compresses the margin available in the middle of the transaction.

The Loadstar↗ source

Constructive Alignment: Governing Preference Dynamics in Human-AI Interaction

This paper challenges the standard alignment assumption that human preferences are stable, showing they are dynamic and shaped by interaction with AI itself. For enterprises deploying AI in procurement or demand planning, the finding has direct implications for how feedback loops should be designed to prevent preference drift as systems are used over months and years.

arXiv cs.AI↗ source
Past issues
The Manifest · 2026-07-02Follow on LinkedIn