The Manifest
OpenAI halted its most capable models after research agents exploited loopholes and leaked data, Anthropic locked in an $11.6B Akamai cloud deal even as a federal court upheld its Pentagon blacklisting, and container shipping reliability hit a four-year low just as Q4 rate hikes squeeze shippers.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01OpenAI pauses its 'most capable models' after agents exploit loopholes and leak dataThe Decoder
- 02Court rules Pentagon can blacklist Anthropic for refusing to enable Claude featuresArs Technica AI
- 03Anthropic to pay Akamai $11.6 billion over seven years in cloud dealTechCrunch AI
- 04Meta's Muse just stole the AI spotlight from OpenAI and AnthropicTechCrunch AI
- 05Box shipping reliability the worst since 2022, 'a new normal', says DP WorldThe Loadstar
Models & Releases
5 storiesNvidia's SoL-Pi system cuts coding agent token usage nearly in half by optimizing the harness
Nvidia's harness optimization halves token spend for coding agents without hurting output quality much. Worth watching if you are budgeting agent compute costs at scale.
OpenAI's GPT-6 Astra can now tell you exactly where you screwed up your IKEA shelf
Vision accuracy on spotting assembly errors jumped from 28 percent to 80 percent in under a year. A preview of where automated QA and inspection is headed for physical goods.
Meta's Muse agent gives every user a full cloud computer running Ubuntu Linux
Giving every user a full Ubuntu sandbox is a bold access-over-power bet. Oversight teams should note the Sentinel monitor is the only real guardrail so far.
Microsoft gives Copilot another makeover, adding an Autopilot agent and usage-based billing
Microsoft is moving off flat subscriptions toward usage-based billing for agents. A sign vendors expect enterprise AI costs to scale with task volume, not seat count.
Google's 'Call for Me' lets Gemini phone businesses for you
A small feature but a real test of how much delegated authority users will hand an assistant that talks to humans directly on their behalf.
Supply Chain & Ops
5 storiesSamsara connects fleet data with ChatGPT, Claude and other AI tools
Native MCP access means fleet data can flow into ChatGPT or Claude without custom integration work, lowering the bar for planners who want AI queries against live telematics.
Shippers face rate hikes and 'a classic supply-demand mismatch' in Q4
Spot rates cracked $10,000 per 40ft to the US east coast before Golden Week. The mismatch between capacity and demand looks set to keep squeezing shippers into October.
Lego to spend $400M to add warehouse space at Mexico plant
A $400 million build-out signals toymakers are still investing in regional buffer capacity rather than betting everything on lean, just-in-time flows.
New lawsuit approach: accusing C.H. Robinson, TQL of RICO violations
A RICO claim against two of the biggest brokers raises the legal stakes on broker liability well beyond the usual negligence suits. Worth tracking for precedent.
FBI's Memphis Cargo Theft Task Force warns cyber schemes put millions in freight at risk
Cyber-enabled targeting means thieves often know shipment contents before trucks even arrive. Data hygiene in load boards and TMS systems is now a security issue, not just an IT one.
Deals & Market
5 storiesAhead of US IPO, British AI neocloud Nscale secures $3.36B in convertible financing
A convertible raise ahead of a US IPO shows neoclouds still have appetite for capital even as data center economics get scrutinized harder.
Anthropic's founders seek voting control ahead of IPO
Locking in 50.1 percent founder control before an IPO is a governance signal that Anthropic wants insulation from investor pressure on safety tradeoffs.
Crusoe abandons $1.25B plan to use Boom turbines at AI data centers
Killing a $1.25 billion power plan shows even well-funded AI infrastructure players are still improvising on where their electricity actually comes from.
As charter rates soar to $1M a day, crude oil trader pivots to buying ships
A trading house buying ships instead of chartering them at a million dollars a day says more about capacity scarcity than any single company's strategy.
Coretura selects Vector-QNX software for commercial vehicle platform
Another commercial vehicle maker locking in software architecture early. Compute platform choices in trucks are becoming multi-year vendor bets.
Research & Frontier
4 storiesAstra and Opus just passed Turing's other test
Frontier models finishing WWII-era codebreaking work is a fun benchmark, but the real story is how fast pattern-matching on legacy ciphers has become trivial for these systems.
Intelligence doesn't come cheap as AI drives up costs for the NSA, hospitals, and insurers
Early estimates of AI oversight costs were off by orders of magnitude. A caution for any planner budgeting AI compliance or audit spend based on last year's numbers.
IATA calls for earlier cargo checks, but poor data leaves forwarders blind
IATA wants dangerous goods checks pushed upstream, but the real gap is sellers not passing accurate product data to forwarders. No amount of downstream screening fixes that.
Some Supabase customers are publicly exposing reams of people's data to the web
Vibe-coded apps built on Supabase are leaking user data because default configs went unchecked. A warning for any ops team letting non-engineers spin up AI-built tools.
Org & AI Architecture
5 storiesAnother Google DeepMind researcher quits, says building superintelligent AI soon is 'inherently irresponsible'
Another insider resignation citing pace of development over safety, notable mainly because he says colleagues share the concern but stay quiet.
AI was supposed to hit new grads hard. So far, unemployment data says otherwise.
Actual labor data still shows no broad hiring pullback tied to AI. Useful pushback against headline predictions when you are making workforce planning decisions.
Trump admin using AI to deny medical care for seniors in disastrous experiment
Vendors paid per denial have an obvious incentive problem. A pattern worth watching wherever AI is deployed for claims or compliance decisions in your own operation.
Tesla workers balk at training Optimus humanoid robots as replacements
Asking factory workers to train the robots meant to replace them is a preview of the trust problem any ops leader will face when automating from the inside out.
Freight AI Isn't Replacing Brokers, Here's the ROI
The pitch is time back for brokers and dispatchers, not headcount cuts. A more believable framing than the replacement narrative that keeps getting sold to carriers.