THE WORLD IS NOT STANDING STILL.RSS
BIG CHANGE.
BIG CHANGE, one tap away
BIG CHANGE, one tap away

Make room for the big picture.

Add BIG CHANGE to your home screen or desktop. It opens in its own window, with the latest stories when you’re online.

iPhone & iPad
Open bigchange.ai in Safari. Tap Share, then Add to Home Screen. Turn on Open as Web App if shown, then tap Add.
Android
Open bigchange.ai in Chrome. In the browser menu, choose Install app or Add to Home screen.
Desktop
In Chrome or Edge, use the install icon in the address bar or the browser’s app menu. In Safari on Mac, choose File → Add to Dock.
All briefings

Follow the change ·

The day,
in perspective.

What happened in AI and robotics — and the change behind it. A reading path through 19 stories from BIG CHANGE.

Ordered using RADAR attention, story opens and likes. Excerpts from our published reporting; open a story for the full context and sources. Dates in UTC.

analysis

Anthropic’s reported founder voting plan would add another layer of control before its IPO

A reported proposal would give Anthropic’s seven co-founders 50.1% of most shareholder votes. Its existing independent Trust has separate board powers, and final IPO rights remain undisclosed.

The big change

What changed: Anthropic has publicly described a governance structure in which its Long-Term Benefit Trust gains authority over board elections. The newly reported proposal would give the founders majority voting power on most other corporate matters, subject to shareholding conditions. Its adoption remains unconfirmed.

Why it matters: A public shareholder’s financial stake and voting influence could differ sharply. The reported design would keep founders influential while preserving a separate route for the Trust to shape the board, where oversight of company leadership sits.

What to watch: The prospectus and governing documents would establish the actual share classes, voting exceptions and thresholds. Those terms would show how much say public investors, employees, founders and the Trust each have after a listing.

Read the full story
—
Back to the briefing
analysis

Open-weight models handled 56% of Vercel’s tokens, but 14% of spend

Vercel's AI Gateway says open-weight models accounted for 56% of August token volume and 14% of list-price-estimated spend in its traffic. The split is not market-wide or an invoice total.

The big change

What changed: Open-weight models moved from a minority to 56% of tokens on Vercel AI Gateway between December and August, while taking 14% of August’s list-price-estimated spend.

Why it matters: Teams choosing models need to separate token demand from spend, then compare quality and full operating cost on the tasks they run. The index provides no task-specific results or savings for an individual team.

What to watch: The practical question is whether a chosen model meets a team’s quality, latency, privacy and operating requirements at its full cost. Token share alone cannot answer that; license terms, hardware and engineering time also matter.

Read the full story
—
Back to the briefing
analysis

DeepSeek details its DSec sandbox system for agent training

A September research report describes DSec, DeepSeek's sandbox platform for agent training, including multiple backends, composable environments and author-reported deployment scale.

The big change

What changed: DeepSeek has documented DSec as a shared platform for agent training and evaluation, with function calls, containers, microVMs and full VMs available through one internal client library.

Why it matters: The report connects agent training to the infrastructure that keeps many task environments running at once. Requests pass through authorization, placement and local admission; layers are composed at creation and image data fetched as needed. Training can pause stateful sandboxes when GPU jobs are preempted.

What to watch: Callers still choose the backend. The paper's production workload measurements cover containers and microVMs, which use different storage paths and incur different resource costs. Its scale figures describe one DSec unit.

Read the full story
—
Back to the briefing
analysis

Australian universities set different rules for AI in marking

Newcastle, Melbourne and UNSW set different boundaries for AI-assisted marking. Their rules also differ on disclosure, opt-outs and academic responsibility.

The big change

What changed: Australian universities are spelling out where generative AI can enter assessment. Newcastle allows its use in marking under review, notice and opt-out conditions; UNSW tells markers, as a rule, not to use AI platforms for marking or grading.

Why it matters: A student can receive AI-assisted feedback without an AI system having final authority over a mark. The difference rests on what the academic actually checks and what the student is told before submitting work.

What to watch: Newcastle and Melbourne require notice and an opt-out route when staff use generative AI in assessment. Their public guidance does not show how often students receive that notice or how staff review the output.

Read the full story
—
Back to the briefing
analysis

GPT-Policy tests context in robot-arm and mobile-exploration tasks

A September preprint reports a three-trial mobile-exploration task alongside robot-arm experiments. It does not verify the separate Reddit claim about GPT-6 Astra and a Unitree G1.

The big change

What changed: GPT-Policy gives a general vision-language model a structured way to turn demonstrations and current camera views into robot-tool requests, then use execution feedback to choose another action without updating the model’s task-specific weights.

Why it matters: The approach separates broad visual reasoning from motion checks and execution. In the authors’ limited trials, added context helped on some tasks, while contact-sensitive actions sometimes needed aligned robot-action references.

What to watch: The paper reports only three trials per condition and describes slow, failure-prone runs, including arm collisions. Reproduction, larger evaluations and independent safety controls matter before these results say much about robots working around people.

Read the full story
—
Back to the briefing
analysis

Six Democrats' AI policies range from state orders to proposed federal laws

Six Democratic officials have taken different routes on AI oversight, data-center construction and worker support. State orders are in effect; federal measures remain proposals.

The big change

What changed: AI policy is taking several forms at once. State governors have issued data-center or oversight orders, while lawmakers have filed bills and a resolution about construction, local control and workers.

Why it matters: A developer in Pennsylvania or Kentucky already faces state-level requirements that a proposed federal bill cannot impose today. Residents, workers and companies need to know which rules apply now and which ideas still require legislation.

What to watch: Pennsylvania's permit process and California's November recommendations will show how existing executive action develops. In Congress, the test is whether the proposed moratorium, community-rights measure or worker fund progresses beyond introduction.

Read the full story
—
Back to the briefing
analysis

Blue Cross estimates $942 million in added costs from complex hospital coding

A new Blue Cross claims analysis estimates $942 million in added hospital payments from more complex coding. It raises questions about AI tools but does not establish their causal share.

The big change

What changed: More inpatient stays were billed in complex categories. BCBSA estimates the shift raised payments across Blue plans by $942 million over two years, with $653 million attributed to additional secondary diagnoses.

Why it matters: Software that searches records and lab results can help hospitals document conditions. In a payment system where some diagnoses change the billing category, documentation can also change the price of a stay. Patients, employers and insurers have an interest in whether the added codes reflect patients' clinical needs.

What to watch: The insurer compared diagnoses with treatments recorded on claims, but did not review each patient's chart or identify which hospitals used which AI tools. Clinical validation would help resolve the dispute over appropriate coding and patient severity.

Read the full story
—
Back to the briefing
analysis

HomeBody connects a humanoid’s spatial memory to reusable robot skills

Caltech and Stanford researchers show a Unitree G1 using a remembered kitchen map and reusable skills. The release has selected demonstrations but no controlled results, full paper or robot code.

The big change

What changed: HomeBody gives a vision-language model access to a spatial record of an explored room and a common interface for robot skills. The model selects a skill and target; local software plans and executes the movement, then reports the outcome.

Why it matters: The model can plan across locations and use feedback without a newly trained action policy for this kitchen. The team’s demonstrations show how memory and existing motor skills can be assembled for longer tasks.

What remains open: The public release shows selected demonstrations, not a controlled success-rate study. A paper and the robot code are not publicly available as of September 27, so the reported performance cannot yet be independently reproduced from the release.

Read the full story
—
Back to the briefing