The Manifest
OpenAI, Anthropic, Google and rival lab heads are publicly coordinating around AI safety pacing (with OpenAI's IPO pushed to 2027) even as GPT-6 Astra and new open-weight agents push capability forward, while freight forwarders face fresh M&A rumors and mid-size carriers get squeezed out of truck financing.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01GPT-6 Astra pilots a surveillance drone and runs a business on its ownThe Decoder
- 02Altman, Musk, and Hassabis back Amodei's call to add independent oversightThe Decoder
- 03Manuport Logistics takeover – rumours buildThe Loadstar
- 04Bank exits squeeze truck financing for mid-size fleetsFreightWaves
- 05Iris-mini and Iris-pro are the strongest open-weight search agents in their classThe Decoder
Models & Releases
4 storiesGPT-6 Astra pilots a surveillance drone and runs a business on its own
OpenAI's model handles autonomous drone piloting and running a vending business with fewer bad-faith moves than Claude Fable 5.1. Worth watching whether these benchmarks translate to warehouse or fleet automation pilots.
Iris-mini and Iris-pro are the strongest open-weight search agents in their class
Open-weight search agents built on Qwen beating benchmarks in their size class matter for teams that want self-hosted research or procurement tools without vendor lock-in.
Elevenlabs makes Music v2.5 available via app and API with free and pro tier options
Licensed-only training and a free tier lower the barrier for teams needing branded audio without licensing risk. Not core operations but worth a look for marketing and internal content teams.
Garry Tan wants US open-weight AI labs to 'distill' frontier models, too
Y Combinator pushing open-weight labs toward distillation is a bet on cheaper, controllable models for buyers wary of depending entirely on closed APIs.
Supply Chain & Ops
4 storiesNSR incumbents double down as newcomers join Arctic shipping push
Rosatom and new entrants adding Arctic-class capacity signals another route competing for cargo away from Suez and Red Sea chokepoints. Planners should weigh seasonal reliability before shifting volume there.
Mexico's Pacific ports race to keep pace with container growth
Manzanillo and Lazaro Cardenas are posting double-digit growth but infrastructure is lagging. Expect congestion risk for anyone routing volume through Mexico to sidestep US-China trade friction.
News in Brief Podcast | Week 37 2026 | EU CBEC Ecommerce Forum and box shipping resilience
Hapag-Lloyd's ZIM bid, renewed Hormuz tensions and container trade holding up despite disruption are the three threads to track this week for rate forecasting.
Borderlands Mexico: Exports of Mexican-made heavy-duty trucks to US roar back
Rebounding truck exports plus a new Texas manufacturing campus show nearshoring capacity is still expanding despite tariff headlines.
Deals & Market
2 storiesManuport Logistics takeover – rumours build
Renewed talk of Maersk circling Euroports' Manuport unit points to more forwarder consolidation. Shippers using independent forwarders should watch for service changes if a major carrier absorbs one.
Bank exits squeeze truck financing for mid-size fleets
Lenders pulling back from mid-size fleet financing after the freight recession means carriers need cleaner books to get approved, a real constraint on capacity recovery heading into peak season.
Research & Frontier
4 storiesLook Before You Leap: Pre-Action Verification for LLM Agents
Proposes a cheap deterministic check before an agent's action executes. Directly relevant for anyone letting agents touch procurement systems or shipment records, where a wrong action can fail silently.
Performance, Efficiency and Collapse -- Advantages and Challenges in Offline Post-training of Code LLMs
Documents how reinforcement learning post-training on code models can degrade performance under certain conditions, a caution for teams building custom coding agents for ERP or WMS integration work.
Efficient AI Model Deployment Using Quantization Analysis Tool
Another edge-deployment optimization paper, useful if you're running models on warehouse handhelds or plant-floor sensors without a cloud round trip.
FINESSE: An Agent-Based Simulator and Benchmark Dataset for Multimodal Financial Event Sequences
A new open benchmark for financial event sequences could help teams building forecasting or fraud-detection agents test against realistic transaction data instead of toy datasets.
Org & AI Architecture
5 storiesAltman, Musk, and Hassabis back Amodei's call to add independent oversight
Three rival lab heads publicly backing independent oversight, plus OpenAI pushing its IPO to 2027, shows safety concerns are now affecting real business timelines, not just messaging.
Sam Altman calls for pacing AI development but promises rapid progress will continue
OpenAI says it now runs safety checks before major training runs and has been talking self-regulation with Anthropic and Google for months. Labs clearly prefer private coordination over waiting on regulators.
What's behind the AI industry's latest warnings of doom?
The renewed existential-risk argument among AI leaders is as much about positioning ahead of policy fights as genuine alarm. Read it skeptically.
Obama urges Democrats to have a 'clear plan' for AI safeguards
Obama pushing Democrats to get concrete on AI policy suggests the issue is moving from lab PR into election-cycle politics, which could reshape the regulatory timeline operators plan around.
Two-year university study finds banning AI from classrooms leaves students worse off
Two years of data show students banned from AI tools finished last both years compared to structured or unguided use. Useful ammunition for ops leaders fighting internal bans on AI copilots.