The Manifest
Hormuz reopens to a cautious trickle of container traffic as US-Iran talks hold, a unanimous Supreme Court ruling strips freight brokers of decades-old liability protection, and Anthropic's Claude Tag signals AI beginning to own institutional knowledge at the Slack layer.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01The Supreme Court and the White House Just Changed Everything for Freight Brokers, NVOCCs, Customs Brokers, Freight Forwarders, and Warehouse OperatorsFreightWaves
- 02Trickle of Box Ships Brave Mines to Pick Their Way Through HormuzThe Loadstar
- 03Claude Tag Embeds Anthropic's AI in Slack, Already Writes 65 Percent of Internal Code, Company SaysThe Decoder
- 04'Collateral Damage?' US Customs Crackdown Alarms Importers and BrokersThe Loadstar
- 05Anthropic Backer Menlo Ventures Raises $3B in New Funds to Back AI Startups Across StagesCrunchbase News
Models & Releases
5 storiesMistral's New OCR Model Beats Competitors in 72 Percent of Blind Test Cases
OCR 4 supports 170 languages, returns bounding boxes and confidence scores per block, and runs in a single self-hosted container. For teams processing invoices, bills of lading, or shipping manifests at scale, this is worth a pilot before the next vendor contract renewal.
OpenAI Says New GPT-5.5-Cyber Outperforms Anthropic's Mythos on Cybersecurity Benchmark
OpenAI is shifting from vulnerability detection toward automatic patching, expanding its partner network to more than 25 security firms. Supply chain teams running vendor portals or EDI integrations should note that AI-assisted patching is arriving faster than most security roadmaps assumed.
Cursor Announces Its Own AI Model, a New Git Platform, and a Mobile App
Building a proprietary model is a bet that deep environment-model integration beats general-purpose assistants. For ops teams building internal tools, the direction narrows the gap between idea and working software without requiring a dedicated engineering team.
Sakana AI's Fugu Orchestrates Multiple LLMs to Match Anthropic's Fable and Mythos Benchmarks
Fugu routes tasks across models in real time rather than committing to one provider. This multi-model orchestration pattern matters for enterprises worried about concentration risk on a single AI vendor, and it shows viable benchmark parity is achievable without a frontier-scale training run.
Datalab Releases lift: A 9B Open-Weights Vision Model That Extracts Structured JSON From PDFs Using Schemas
lift hits 90.2% field accuracy and returns null instead of hallucinating absent fields. That abstention behavior is what makes it actually deployable in document extraction workflows where a wrong value is worse than a missing one.
Supply Chain & Ops
5 storiesThe Supreme Court and the White House Just Changed Everything for Freight Brokers, NVOCCs, Customs Brokers, Freight Forwarders, and Warehouse Operators
The May 22 ruling in Montgomery v. Caribe Transport II removes a liability shield brokers have relied on for decades, arriving alongside a White House executive order raising customs penalties. Carrier vetting and compliance processes that were nice-to-have are now legally load-bearing.
Trickle of Box Ships Brave Mines to Pick Their Way Through Hormuz
A 60-day US-Iran MoU has reopened the strait, but mine risk means transits are slow and selective. Planners should extend lead time buffers on Persian Gulf origin cargo until a permanent deal is in place.
Gulf Carriers Drive Recovery, Yet Airfreight Rates Stay Stubbornly High
Gulf airlines are rebuilding capacity quickly, but rates are not following freight volumes down despite easing geopolitical tension and lower fuel. If your air cargo contracts come up for renewal in the next 60 days, pushing for spot indexing rather than fixed rates may be the better hedge.
US Revokes 20,000 Visas for Mexican Truckers as Cabotage Crackdown Expands
Losing 20,000 cross-border drivers since April 2025 will show up in Mexico-US lane capacity and drayage wait times. If you run nearshored production in Mexico, verify your carrier network still has compliant coverage before the next production cycle.
Kraft Heinz Merges Procurement and Supply Chain Units
Naming a single officer over both procurement and supply chain, effective July 1, signals that Kraft Heinz wants sourcing and execution decisions made in one conversation. Expect similar consolidations at other large CPG firms navigating continued tariff volatility.
Deals & Market
4 storiesAnthropic Backer Menlo Ventures Raises $3B in New Funds to Back AI Startups Across Stages
Menlo's largest fund in 50 years targets enterprise AI from seed through growth, meaning more capital is available to bring vertical tools, including supply chain applications, from pilot to full deployment faster than prior cycles.
FedEx Boosts Revenue Behind Premium Parcel, Freight Volumes
FedEx is winning on yield rather than volume as it concentrates on high-margin freight. Shippers should expect continued pricing discipline on time-definite services, with little sign of softening from the carrier.
Oracle's 21,000 Layoffs Help Drive Its Debt-Fueled AI Investments
Oracle is converting headcount savings into data center buildout at scale. For ops teams evaluating Oracle's supply chain modules, the product roadmap bet on AI infrastructure is real, but near-term support headcount is worth scrutinizing in contract negotiations.
AI First: How Fura Grew 800% During Freight Slump
An AI-first brokerage turning a $150K loss into $1M profit during the freight recession is a proof point worth examining closely. The model likely succeeds on operational cost structure rather than anything algorithmic, so the question is which specific cost lines it eliminated.
Research & Frontier
4 storiesDFlash Speculative Decoding Drafts Whole Token Blocks in Parallel for Up to 15x Higher Throughput on NVIDIA Blackwell
DFlash replaces sequential token drafting with block-parallel diffusion, reporting 15x throughput on Blackwell at interactive latency targets. This class of inference speedup is what will eventually make real-time AI planning assistants economical to run at enterprise scale.
RIFT-Bench: Dynamic Red-Teaming for Agentic AI Systems
RIFT-Bench tests agentic systems across heterogeneous environments rather than fixed domains, which is a closer simulation of how supply chain agents will actually be deployed. Teams evaluating AI automation vendors should ask for red-team results against comparable benchmarks before signing.
Neuro-Symbolic Drive: Rule-Grounded Faithful Reasoning for Driving VLAs
The paper grounds autonomous vehicle reasoning in formal traffic rules so the model's stated rationale and its actual motion planning stay in sync. The same constraint-grounding approach is directly relevant to any agentic system where auditability of decisions is a compliance requirement.
Reinforcement Learning Towards Broadly and Persistently Beneficial Models
The paper studies whether RL alignment generalizes beyond training domains, focusing on reward hacking and deception in high-stakes settings. For teams deploying RL-trained agents in procurement or demand planning, the findings on out-of-distribution behavior are worth reviewing before scaling.
Org & AI Architecture
4 storiesAnthropic's Claude Tag Is Learning Your Company, One Slack Message at a Time
Beyond productivity, Claude Tag is a play to own institutional knowledge at the Slack layer. Organizations should think carefully about what proprietary process knowledge is being ingested and whether existing data governance agreements actually cover it.
Pangram CEO Says Language Models Give Themselves Away by Making the Same Arguments
LLMs cluster around the same reasoning paths even when asked for diverse inputs, which matters for any use case that relies on AI-generated scenario planning or risk analysis. Running multiple models or architectures in parallel may surface more genuine variation than prompting harder.
How to Burst the AI Bubble: Strike at Its Roots
Cory Doctorow's argument centers on data access and concentration of AI infrastructure ownership rather than the technology itself. A useful counterweight to read before committing to long-term AI vendor contracts where switching costs are high.
Lands' End Warehouse System Rollout Spurs Temporary Backlog
A WMS go-live that caused a fulfillment backlog, then resolved within days, is a reminder that cutover planning and sequencing matter as much as the system itself. Tracking time-to-fulfill as the recovery metric, as the CFO did here, is the right approach to validate the go-live is actually stable.