The Manifest
OpenAI is deliberately slowing model releases over cyberattack risk, Anthropic's revenue hit a $65 billion annualized rate ahead of a possible IPO, and ocean freight rates spiked on carrier capacity discipline while Walmart tightened its grip on suppliers' first-mile logistics.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01OpenAI says it's 'pacing model development' as AI cybersecurity risks grow too dangerousThe Decoder
- 02Anthropic increases revenue sevenfold, hits annualized rate above $65 billionThe Decoder
- 03Ocean Freight Rates Surge +400% — But Demand Isn't Driving ItFreightWaves
- 04Etched's valuation doubles to $21B in a monthTechCrunch AI
- 05Walmart tightens grip on the first mile – suppliers' optionality is the priceThe Loadstar
Models & Releases
5 storiesOpenAI says it's 'pacing model development' as AI cybersecurity risks grow too dangerous
OpenAI is deliberately slowing releases because its next model may cross into dangerous cyberattack territory. Teams running AI-assisted security tooling should watch for tighter API guardrails and slower feature rollouts ahead.
GLM-5.3 Artificial Analysis Benchmarks
The new GLM release lands high on Artificial Analysis rankings, another sign that open-weight Chinese models are closing the gap with frontier US labs. Worth tracking if you're evaluating cheaper inference options.
Cartesia Ships Sonic-3.6: A Streaming TTS Model That Now Leads Both Artificial Analysis Speech Arenas
A non-transformer TTS model now tops both Artificial Analysis speech leaderboards with sub-90ms latency. Relevant for anyone building voice agents into customer service or warehouse floor interfaces.
NVIDIA Releases TensorRT Model Connect in Public Preview: Hugging Face Checkpoint to Native C++ Inference in Two Commands
NVIDIA cuts the path from a Hugging Face checkpoint to production C++ inference down to two commands. Useful for ops teams trying to deploy models without a dedicated ML engineering bench.
Meet SAM (Sovereign Agent Mesh): A Zero-Config, Zero-Trust P2P Network for AI Agents
Google open-sourced a zero-trust network that lets AI agents call each other's tools across cloud and on-prem environments. Early infrastructure worth watching for anyone planning multi-agent workflows across a supply chain stack.
Supply Chain & Ops
5 storiesOcean Freight Rates Surge +400% — But Demand Isn't Driving It
Spot rates are spiking on blank sailings and carrier capacity discipline, not actual demand growth. Treat this as a pricing power story rather than a volume signal when budgeting Q4 ocean freight.
Walmart tightens grip on the first mile – suppliers' optionality is the price
Walmart's Prepaid Consolidation program looks like a cost win up front, but suppliers who join lose leverage to shop other carriers later. Read the contract terms closely before opting in.
Customs to fight this growing trade fraud with AI 'detective border'
A new White House report flags transshipment fraud as a growing target for AI-driven customs screening. Importers relying on gray-zone routing should expect more scrutiny at entry points.
Honeywell Aerospace targets supplier bottlenecks to boost output
A meaningful share of Honeywell's 3,000 suppliers underperformed in H1. A reminder that even well-resourced OEMs are still fighting supplier quality and capacity issues two years into the recovery narrative.
Panama Canal continues draft restrictions, ocean carriers up fees
MSC and CMA CGM are layering new surcharges on top of ongoing draft restrictions. Build canal risk premiums into transpacific and Caribbean routing costs now rather than after peak season.
Deals & Market
5 storiesAnthropic increases revenue sevenfold, hits annualized rate above $65 billion
Annualized revenue above $65 billion and talk of a fall IPO at a $1 trillion valuation would put Anthropic ahead of OpenAI to market. Enterprise buyers should watch for what that scale does to pricing and roadmap stability.
Etched's valuation doubles to $21B in a month
A Jane Street-backed round doubling Etched's valuation in a month shows how much capital is chasing specialized inference chips. Worth watching if you're weighing custom silicon against general GPU capacity for AI workloads.
Relativity Networks raises $22 million to bring a faster kind of fiber to data centers
Hollow-core fiber promising 30% faster data transmission targets data center interconnects. A niche but real lever for anyone tracking AI infrastructure costs that eventually show up in cloud pricing.
DOJ probes Andreessen Horowitz over partners sitting on competing AI boards
An antitrust probe into a16z partners sitting on competing AI company boards is a reminder that concentrated VC influence over data and AI infrastructure is starting to draw regulatory attention.
Perplexity's free AI offer left it with millions more users in India
Revenue rose about 60% even after the free promotional tier ended. Evidence that aggressive market-entry pricing can convert into paying users instead of just burning cash.
Research & Frontier
5 storiesRuntime Governance for Agentic AI: Action-Boundary Control with Trusted Provenance and Fail-Closed Execution
A new framework adds an actual execution boundary around agent tool calls instead of relying on prompt-level rules. Exactly the kind of control layer ops teams need before letting agents touch live systems.
The Price of Thinking: Reasoning Effort as a Model-Specific API Contract
The study treats reasoning-effort settings as part of the API contract itself, a useful frame for procurement teams negotiating LLM contracts who need to know what they're actually paying for beyond the model name.
AI systems quietly drop user instructions when they compress context
Long-running agents lose an average of 83 percent of user rules, like never sending emails without approval, when they compress context. A real risk for anyone deploying agents on multi-day workflows without a persistence layer.
New benchmark ranks search APIs for AI agents on quality, cost, and speed
Artificial Analysis's Search Index gives buyers a way to compare search API providers before wiring one into an agent pipeline. Worth checking before defaulting to whatever your model vendor bundles.
Microsoft Copilot reveals secret input that allowed it to be hacked
Researchers found a hidden parameter that let attackers steal credentials through a crafted link. A reminder to audit any AI assistant with access to internal systems before treating it as trusted infrastructure.
Org & AI Architecture
5 storiesAnthropic CEO says AI centralizes by nature and open models just shift power to whoever owns the chips
The public fight over whether AI regulation entrenches incumbents or checks corporate power isn't just Silicon Valley noise. It will shape how much control enterprises retain over the models they depend on.
As AI beats doctors, regulators shouldn't force a human into the loop, JAMA piece says
The evidence base is still mostly simulations, not real patient outcomes. Treat this argument as a preview of a fight that will eventually hit AI-assisted decisions in operations and quality control too.
OpenAI launches a safer ChatGPT for teens — years after teens started using it
Age-gating and parental controls arrive years after teens were already using the product at scale. Worth remembering as your own org rolls out AI tools to employees without guardrails in place first.
Glossier supply chain leader champions self mastery, stronger representation
A candid look at representation in supply chain leadership from Glossier's VP of Supply Chain and Operations. A useful read for anyone building a more diverse ops leadership pipeline.
3 ways Kimberly-Clark is using technology to boost supply chain efficiency
New procurement software, AI agents and skill-based training are the three levers Kimberly-Clark is pulling at once. A reasonable template for CPG ops leaders scoping their own AI adoption roadmap.