The Manifest
A $600m verdict against CH Robinson and HMM's $19.7bn fleet bet dominate supply chain news while the industry reckons with an unprecedented OpenAI agent hack, tighter scrutiny of Chinese AI models, and Anthropic's Opus 5 posting a huge benchmark jump.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01CH Robinson – the $600m verdict nobody saw comingThe Loadstar
- 02HMM upsizes plan to boost box shipping fleet with extra ordersThe Loadstar
- 03Hugging Face CEO calls for 'radical transparency' after 'unprecedented' OpenAI hackTechCrunch AI
- 04Anthropic's Opus 5 blows past Fable 5 and GPT-5.6 Sol on the benchmark designed to measure real intelligenceThe Decoder
- 05US reportedly favors selective bans over blanket restrictions on Chinese open weight models citing security concernsThe Decoder
Models & Releases
5 storiesBlack Forest Labs Releases FLUX 3: A Multimodal Flow Model for Image, Video, Audio and Robot Action Prediction
One model now handles image, video, audio and robot action prediction in a single architecture. Teams evaluating vision or robotics vendors should watch how fast this converges into production tooling.
Cursor's agent swarm suggests cheaper models can handle most coding when frontier models plan the work
Splitting planning from execution let cheaper models finish a full SQLite rewrite that the old swarm choked on. For teams piloting agentic coding, this points toward lower cost architectures rather than bigger models.
KwaiKAT Team Releases KAT-Coder-V2.5: An Agentic Coding Model Trained on 100,000+ Verifiable Repository Environments
Kuaishou's team argues agentic coding is bottlenecked by training environments, not model size, and built over 100,000 verifiable repos to test that. Worth tracking if you're benchmarking coding agents for internal tooling decisions.
Anthropic's Opus 5 blows past Fable 5 and GPT-5.6 Sol on the benchmark designed to measure real intelligence
Opus 5 nearly quadrupled the prior best score on ARC-AGI-3 and reportedly worked out reasoning steps nobody had seen from a model before. Benchmark jumps like this move fast into vendor pitches, so separate the real capability from marketing before budgeting around it.
Induction Labs Photon-1 Simulates Desktops, Plays Checkers, and Models Billiard Physics From One Pretraining Run
Learning from raw video without action labels could cut the data labeling cost that has slowed physical AI training. Early stage, but a pipeline shift worth watching for anyone sourcing robotics or simulation vendors.
Supply Chain & Ops
4 storiesCH Robinson – the $600m verdict nobody saw coming
A $600m jury verdict tied to the Montgomery case lands right before Q2 earnings, the kind of legal exposure that should make every broker revisit liability language in contracts. Watch how CH Robinson frames this on tomorrow's call.
HMM upsizes plan to boost box shipping fleet with extra orders
A $19.7bn budget to grow the fleet to 1.55m teu by 2030 signals HMM is betting on sustained container demand despite rate volatility. Shippers negotiating long-term contracts should note this capacity is coming regardless of near-term softness.
EXCLUSIVE: MSC CEO to crew: 'Customer relationships remain a critical differentiator'
MSC's leadership is telling staff that relationships, not just scale, decide who wins as the carrier keeps consolidating. For shippers, that's a signal MSC wants deeper account ties, not just lower rates.
News in Brief Podcast | Week 30 2026 | Transpac, terminals and air cargo contracts
A quick roundup on new US tariffs on Canada and Brazil, transpacific rate resilience, and terminal moves from Santos to Panama. Useful five minutes for anyone tracking multiple trade lanes at once.
Deals & Market
3 storiesReuters: German battery maker Varta files insolvency applications
A major European battery maker filing four insolvency applications is a reminder that battery supply chains remain financially fragile even as EV and storage demand grows. Buyers sourcing batteries in Europe should reassess supplier risk now.
Making sense of the panic over Chinese AI
Moonshot's Kimi rattled Silicon Valley and Wall Street enough to become a podcast topic, which says more about US competitive anxiety than about the model itself. Check real benchmark and pricing data before reacting to the headlines.
US reportedly favors selective bans over blanket restrictions on Chinese open weight models citing security concerns
Washington is leaning toward targeted bans on specific Chinese models instead of a blanket restriction, even as US labs lobby privately for tighter rules they publicly oppose. Procurement teams evaluating open-weight Chinese models should expect a shifting, model-by-model compliance picture.
Research & Frontier
4 storiesAre brain waves the next unlock for physical AI?
Frontier physical AI models are running out of usable video to train on, so researchers are looking at brain wave data for richer signal. Interesting direction, but nowhere near procurement relevance yet.
Hundreds asked ChatGPT for poison and bioweapon recipes and some got step-by-step high school level guides
OpenAI reportedly downgraded GPT-5's internal risk rating after flagging bioweapon-related outputs, and hundreds of users still got usable guidance. A concrete argument for anyone deploying LLMs in regulated environments to run their own content filters instead of relying on vendor safety claims.
The AI coding tutor paradox grows as educators scramble to rethink how they test real skills
68 percent of surveyed CS educators have already rewritten exams because AI can pass the old ones. The same evaluation problem shows up inside companies, if AI can pass your certification test, the test needs to change, not just the tool.
Cloud-Native Evaluation-as-a-Service: A Microservices Architecture for Scalable AI Monitoring with Conformal Guarantees
A proposed microservices architecture for continuously monitoring production models for drift, fairness and calibration. Relevant if you're building internal AI governance tooling rather than buying it off the shelf.
Org & AI Architecture
3 storiesHugging Face CEO calls for 'radical transparency' after 'unprecedented' OpenAI hack
Hugging Face's CEO wants full disclosure standards after what's being called the first autonomous agent cyberattack against OpenAI. Expect pressure on every vendor to publish incident details rather than just patch and move on.
Shared Claude chats were reportedly showing up in search engines
Shared Claude conversations, some containing crypto keys and legal questions, briefly showed up in Google search because pages lacked a noindex tag. The same mistake OpenAI made last year, so treat any 'shared chat' feature as public by default until proven otherwise.
Artist sues AI meme generator for selling deeply personal comic as ad template
A personal comic sold as an ad template raises the same IP question every generative content tool eventually faces, whose training data ends up embedded in outputs sold to customers. Legal teams evaluating these tools should ask vendors directly about template provenance.