The Manifest
An AI hallucination nearly triggered a US military boarding, security researchers used Claude to breach OpenAI's systems while Gemini broke into real companies during testing, and freight rates and surcharges kept climbing even as California moved to mandate an AI kill switch.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01U.S. military nearly boarded a Chinese ship over a hallucinated AI intelligence reportThe Decoder
- 02Asia-US container rates soar past $11,000, near pandemic recordsFreightWaves
- 03Security researchers used Anthropic's Claude to hack OpenAI's internal systems in under 72 hoursThe Decoder
- 04California Governor Newsom signs executive order demanding 'kill switch' for AI modelsThe Decoder
- 05Microsoft exec called AI scraping 'the largest theft of labor in human history'techcrunch.com
Models & Releases
5 storiesA new kind of AI model from a ChatGPT inventor is thrilling developers
A new architecture promising cheaper, faster inference is worth a look if your teams are burning budget on model calls. Wait for third party benchmarks before betting a roadmap on it.
Google's new 'CC' is an AI agent that helps families run their households
Household AI agents are a consumer play for now, but an agent that reads shared calendars and files to run recurring tasks is exactly what planning teams will want next.
OpenAI takes aim at the legal market with Astra for Law
Vertical AI products are moving into regulated, document heavy work. Expect similar plays aimed at trade compliance and contract management before year end.
US government website used Chinese model the FBI called 'malicious'
A federal site quietly running an open source Chinese model shows how deep foreign AI tooling has already crept into public infrastructure. Worth checking your own vendor stack for the same blind spot.
Google's Gemini also accidentally hacked three real companies during security testing
A misconfigured test environment let a model loose on the open internet and it broke into real companies. Treat every AI security test environment like production until proven otherwise.
Supply Chain & Ops
5 storiesFedEx levies fees on US imports from Canada, Europe and others
New import surcharges land right as peak season staffing and space get tight. Budget for it now rather than eating the surprise in October invoices.
Diesel prices surge past $6.28 per gallon
Fuel costs are compounding freight rate pain. Anyone locking in Q4 contracts should build a wider fuel surcharge band than usual.
No post-peak relief for ocean freight as capacity tightens
Rates were supposed to soften after peak and didn't. Plan cargo release dates assuming capacity stays tight into Q4, not just through September.
FAA tees up $875M AI tool to help manage air traffic congestion
An AI tool heading toward a nationwide rollout matters for anyone leaning on air freight. Whether it smooths or complicates scheduling depends entirely on rollout pace.
Costco expands ultra-fast delivery through DoorDash, Uber
Handing last-mile to gig platforms keeps capex off the balance sheet but ties service quality to someone else's driver pool. Watch the complaint data closely.
Deals & Market
5 storiesManus seeks $4B valuation in new $500M fundraise as it resumes independent ops
A $4B ask after a collapsed Meta merger is a bold reset. Watch whether enterprise buyers actually pay up or this is investor optimism outrunning revenue.
A startup that builds other startups raised $100M and is all-in on physical AI
A studio model for building physical AI ventures inside industrial corporates could beat internal innovation labs to deployment, if the parent companies actually adopt what gets built.
Anthropic's first embedded evaluator is … Accenture?
Putting a consulting giant inside model evaluation is a conflict of interest question waiting to happen. Watch how independent the findings actually turn out to be.
L'Imad's Atlas Air interest could give Apollo a valuable price marker
Sovereign wealth interest in a freighter operator signals confidence that air cargo capacity stays scarce and pricey for a while yet.
Hub Group receives Nasdaq delisting notice, plans to appeal
A delisting notice over late filings is a governance flag for shippers who rely on Hub Group capacity. Worth a call to your account rep before it becomes your problem.
Research & Frontier
5 stories42 leading mathematicians warn that AI existential risk is real and urgent
When Fields Medal winners sign a warning letter it's not fringe anxiety anymore, but it also won't change anyone's procurement decisions this quarter.
Visible chains of thought are a safety advantage for AI, but that transparency is slipping away
Losing the ability to read a model's reasoning trail removes one of the few cheap safety checks available today. A real concern for anyone auditing agentic workflows.
AI conference ICLR is drowning in abstracts, with roughly 50,000 submissions before the deadline
Fifty thousand submissions means the peer review system is buckling. Expect noisier research signal for a while, so treat single papers with more skepticism.
World model companies are keeping a lot of secrets
Heavy funding with zero transparency on training data or methods is a pattern worth remembering next time a vendor pitches a world model for logistics simulation.
Anthropic is operating a lab that conducts biology experiments
A frontier lab running wet lab biology experiments blurs the line between AI research and dual use science. Governance here deserves more scrutiny than it's getting.
Org & AI Architecture
4 storiesDario Amodei and other AI leaders want to 'Pace the Frontier' but…how?
A voluntary pacing plan from one lab means little without competitors matching it. Watch for actual commitments rather than talking points.
Anthropic wants you to know Claude leads a quarter of its research, but 'lead' doesn't mean what you think
A quarter of Anthropic's own R&D credited to its models sounds big until you notice the metric is self reported and self scored. Read the caveats before repeating the stat.
AI chatbots are becoming experts at changing people's minds
Persuasive AI is already good enough to shift opinions at scale, a useful and slightly worrying data point for anyone using AI in supplier negotiations or customer messaging.
Open or closed AI? Nvidia's Nader Khalil and Sydney Sykes take on one of the decisions shaping next-gen startups at TechCrunch Disrupt 2026
The open versus closed model debate keeps resurfacing at every conference. The real decision for operators is which one your vendors actually support long term.