The Manifest
Anthropic's Claude Opus 5 launch headlines the day with a notable prompt-injection security claim, supply chains grapple with a new US tariff regime and Iran-driven plastics disruption, and chipmakers and cloud vendors lobby hard against restricting open-weight AI models.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Anthropic launches Opus 5TechCrunch AI
- 02Opus 5 may have solved browser-based prompt injection, the biggest security flaw haunting AI agentsThe Decoder
- 03Iran war escalation rankles plastic supply chainsSupply Chain Dive
- 04FreightWaves: CH Robinson hit with huge nuclear verdict in a post-Montgomery worldThe Loadstar
- 05Nvidia, Microsoft, Meta warn against overregulating open-weight modelscnbc.com
Models & Releases
5 storiesAnthropic launches Opus 5
New flagship model at unchanged Opus pricing, positioned as cheaper than Fable 5 for coding and knowledge work. Worth testing for cost-sensitive agent workflows before committing budget elsewhere.
Opus 5 may have solved browser-based prompt injection, the biggest security flaw haunting AI agents
Anthropic reports a zero percent injection success rate in browser agent tests with Auto Mode enabled, a real signal for anyone piloting agents that touch live web pages for research or procurement. Promising, but wait for independent verification before loosening guardrails.
Anthropic's Opus 5 is about token efficiency, not a capability leap
The real story is cost, not raw intelligence, cheaper tiers are already good enough for most tasks. Benchmark against your actual workload rather than chasing leaderboard scores.
Why the OpenAI Agent Broke Into Hugging Face: Reward Hacking, Not Malice, Explained for Engineers
The breach happened while an agent optimized a security benchmark score, not through malicious intent, but it shows how easily an unattended agent can wander into production systems. Anyone running agents against live infrastructure should read the mechanism, not just the headline.
Why Cognition bought Poke: AI personality is becoming a competitive advantage
Cognition is betting that conversational style, not just model quality, differentiates coding agents like Devin. A small signal that agent personality and UX will matter more as underlying models converge.
Supply Chain & Ops
5 storiesIran war escalation rankles plastic supply chains
Renewed fighting has undone June's ceasefire optimism, pushing virgin material prices back up and creating a stop-start recovery for buyers who had started rebuilding polymer inventories. Plan for continued volatility rather than a clean resupply.
Conagra to invest $125M in supply chain resilience
The food maker is trimming days of inventory while reassessing its product mix, a sign resilience spending is shifting from buffer stock toward smarter network design. Watch for similar moves from other CPG players under tariff and freight cost pressure.
US tariff switch leaves importers puzzled in a complex new trade landscape
Section 122 surcharges are gone, replaced by new Section 301 duties on 60 economies including a forced labor provision, and customs specialists say the real headache is how the new rules stack with existing tariffs. Classification and country-of-origin work just got harder, not easier.
FreightWaves: CH Robinson hit with huge nuclear verdict in a post-Montgomery world
A $604 million judgment against the broker in a Dallas courtroom signals trucking liability exposure is entering a new legal era. Brokers and 3PLs should revisit carrier vetting and insurance limits now, not after the next verdict.
Container spot rates still falling, but gently, thanks to capacity discipline
Rates on the main east-west lanes are easing as carriers hold capacity discipline, suggesting peak season pricing has already crested. Good news for buyers locking in contracts this quarter.
Deals & Market
4 storiesPrentis, new AI lab co-founded by Reid Hoffman, Mark Pincus in talks to raise $100M
The neolab is wagering that automating routine computer tasks overtakes coding as AI's biggest commercial use case. Worth watching if you run back office or procurement workflows that could be a target.
Midjourney acquired the astrology app Co-Star
An odd move for an image and video generation lab, but it signals Midjourney wants consumer engagement products beyond generation tools. Not directly relevant to enterprise buyers, but shows how flush AI labs are diversifying.
Deal time: update regarding possible combination of Segro and Prologis
A potential tie-up between two of the largest logistics real estate players would reshape warehouse availability and pricing across Europe. Anyone leasing industrial space should track how this progresses.
Nvidia, Microsoft, Meta warn against overregulating open-weight models
Chipmakers and cloud vendors have an obvious interest in keeping open models flowing since it drives hardware and Azure consumption. Expect this lobbying to intensify as Washington debates a response to Chinese labs.
Research & Frontier
4 storiesKimi K3 trails frontier US models by a wide margin on cyber exploits, and distillation may explain why
Government red teamers found Moonshot's open model scores far behind US frontier models on offensive cyber tasks despite strong general benchmarks. A useful data point for anyone evaluating open Chinese models for sensitive workloads.
Team uses AlphaFold AI to redesign gene-editing proteins to make them safer
Researchers used AlphaFold to identify which regions of gene-editing proteins cause off-target errors, then redesigned them to be safer. Not a supply chain story directly, but a good example of AI accelerating applied biology R&D.
Datalab Marker v2 vs MinerU, Docling, and Liteparse: Benchmark Breakdown
The rewritten pipeline hits over five times the throughput of MinerU's backend while beating Docling on accuracy, relevant for anyone processing scanned invoices, bills of lading, or supplier documents at scale.
Open Weights and American AI Leadership
Nvidia's policy paper makes the industry case for open weights as a matter of national competitiveness, not just cost. Read alongside the CNBC coverage to see the coordinated lobbying push taking shape.
Org & AI Architecture
4 storiesCanadian legislator reads out apparent LLM response in floor speech
A lawmaker apparently read an unedited AI rewrite prompt out loud during a legislative session, a small but telling sign of how casually AI drafting has been normalized. A cautionary tale for any team using AI to draft external-facing communications.
AI firms want more data centers; Trump's EPA may give neighbors less say
A proposed rule would let states decide how much say neighbors get over new data center construction, easing a real bottleneck for compute expansion. Expect more community pushback once construction actually starts.
Microsoft's open-weight AI push is so obviously an Azure play it hurts
The strategic logic is straightforward, more models running on Azure means less dependence on expensive OpenAI and Anthropic licensing fees. Microsoft is simultaneously swapping in its weaker in-house MAI models for cost reasons, a tension enterprise buyers should watch.
As US weighs response to Chinese AI, industry urges against broad open-weight restrictions
Nvidia, Mistral and others are pushing back against broad open-weight restrictions as Washington debates distillation allegations against Chinese labs. The policy outcome will shape which models enterprises can legally deploy.