The Manifest
AI safety warnings went mainstream and landed a seat on OpenAI's board just as Trump escalated the Canada trade war and typhoon congestion squeezed Asian freight capacity, while data shows top AI spenders are already pulling back on per-employee costs.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01AI safety panic goes mainstream after Anthropic researcher's warnings land on CNN and Fox NewsThe Decoder
- 02Trump escalates Canada trade war with import bans, federal contract restrictionsFreightWaves
- 03OpenAI adds a prominent AI doomer to its board of directorsTechCrunch AI
- 04Typhoon-related port congestion keeps intra-Asia rates high and capacity tightThe Loadstar
- 05Top AI spenders cut per-employee costs by nearly 10 percent in AugustThe Decoder
Models & Releases
5 storiesDeepSeek AI Released DeepSeek-V4.1-Flash with 1M Context, FP4 KV Cache, and Cross-Layer Attention Reuse
For planners weighing model costs, the 1M token context and cheaper serving stack matter more than benchmark scores, this is built squarely for long-running agent workloads that chew through input tokens.
LandingAI Releases Agentic Document Extraction Gen2 with DPT-3 Pro and DPT-3 Verity
Document extraction upgrades like this quietly speed up invoice and customs paperwork processing, but Gen1 code breaks on migration so test before renewing that contract.
Google Open-Sources Mantis: A Modular Skills Toolkit That Lets Coding Agents Find, Reproduce and Patch Vulnerabilities
Open source security tooling for coding agents lowers the bar for teams patching vendor software, relevant as more ops systems ship with AI-written code underneath.
Suno launches v6 music models built with Warner, BMG, and Believe
Music generation isn't core to supply chain, but Suno cutting licensing deals instead of getting sued signals a template other content-heavy AI vendors may follow when facing rights holders.
ChatGPT Images 2.5: Faster, more precise, but not the same for everyone
Incremental image model update, worth noting mainly because OpenAI still isn't transparent about which users get which model behind the scenes.
Supply Chain & Ops
5 storiesBelgium welcomes new ecommerce fee and steps up war on non-compliance
The EU's 3 euro parcel fee has already cut simplified customs declarations in Belgium roughly in half year on year, a preview of the compliance cost any cross border seller needs to budget into landed cost models.
FMCSA suspends USDOT deactivations for missed biennial updates during MOTUS rollout
Carriers and brokers get a grace period on registration deactivations during the systems rollout, small relief but worth flagging to compliance teams tracking USDOT status.
Boston Scientific begins to restore shipping after cyberattack
A device maker is still working through order and shipping backlogs weeks after an attack hit manufacturing and order processing, a reminder that cyber resilience belongs in continuity planning, not just IT.
Amazon Cargo Jet Crash: What Went Wrong in Miami
NTSB findings point to a long landing and possible full thrust before the plane left the runway, anyone leasing converted freighters should watch what this investigation says about maintenance and braking systems.
Freight Audit Is Broken: Real-Time AI Fixes It
Continuous audit that flags billing errors as shipments move beats a monthly invoice review, this is the kind of narrow AI application that pays for itself in disputed charges alone.
Deals & Market
5 storiesAI research startup Listen Labs scrubbed a $1.5B funding round for Salesforce talks
Walking from a signed term sheet to chase a strategic deal says more about enterprise AI distribution channels than about valuation, watch who else takes that trade over a straight VC check.
Freightos founder Zvi Schreiber turns activist in battle to restore growth
A founder going public against his own board is rare in freight tech, and it puts real pressure on Freightos to show growth numbers rather than more strategy slides.
Geis Group acquires Krage & Gerloff Logistik
Consolidation among mid-size European road carriers continues as owners retire, buyers with balance sheet capacity are picking up regional networks at reasonable prices.
Amazon projects reduced reliance on USPS, UPS for delivery: report
If Amazon really delivers over 86 percent of its own volume next year, USPS and UPS lose a meaningful revenue base and every other shipper gains more leverage negotiating with them.
AWS is using Qualcomm for AI inference while Qualcomm uses AWS Bedrock to design the chips
Cloud providers designing custom inference silicon while also selling design tools back to the chipmaker shows how tightly the compute supply chain has knotted itself together.
Research & Frontier
5 storiesDeepmind's AlphaGenome Atlas maps every possible DNA change in the human genome
A petabyte scale map scoring nearly every possible single letter genome change is a research tool today, but the brute force scoring approach is worth watching for other combinatorial problems outside biology.
OpenAI's millennium proof dispute raises the question of whether researchers can trust AI labs
A fight over credit for an AI generated math proof raises real questions about trusting vendor claims of scientific breakthroughs, read the underlying paper before repeating the headline.
AhaBench: Do Agents Learn from Prior Experience? A Benchmark for Long-Horizon Continual Learning
Most agent benchmarks reset after one task, this one checks whether an agent actually improves across a session, closer to how a real workflow assistant needs to behave.
When Does Memory Help? A Cost-Aware Evaluation of Long-Term Memory in Tool-Using LLM Agents
Useful if you're evaluating agent vendors, it tests whether remembered facts actually change what the agent does, not just whether it can recall them on request.
Damage-Aware Bandit Pruning for Vision and Language Transformers
A cheaper way to prune vision and language transformers without wrecking accuracy, relevant for anyone trying to run capable models on smaller hardware budgets.
Org & AI Architecture
5 storiesAnthropic built an economic model that frames its CEO's bleakest job forecasts as an outlier scenario
Anthropic's own scenario modeling puts its CEO's most dire unemployment forecast in the extreme tail, useful context next time a lab executive predicts mass job losses.
Anthropic researcher quits with a warning: Self-improving AI could 'kill us all'
A departing researcher warning that self-improving AI could kill everyone is now hitting cable news, treat it as a data point on internal lab tension rather than a forecast to plan around.
Man told ChatGPT he was feeling delusional. ChatGPT insisted he was Jesus.
A lawsuit over a chatbot reinforcing a user's delusion previews the liability exposure any company deploying conversational AI in customer-facing roles should be tracking.
Apple Watch's new AI features are normalizing the idea that technology is always listening
Ambient transcription features are shifting what always on means for wearables, procurement and IT teams issuing these devices need a policy before employees start wearing microphones by default.
ControlAI's Connor Leahy on why superintelligence is 'not a weapon, it's an adversary'
A useful counterpoint to lab hype, framing superintelligence as an adversary rather than a tool is a debate worth having internally before pinning strategy to any single frontier model roadmap.