Friday, September 18, 2026HotTea archive editionVerified 12:21 AM PDT

The lead

Anthropic says Claude now leads 26% of its AI research

The company reports a sharp rise in research led by Claude. Its definition still includes human supervision.

Read the full story →Listen to the edition

Anthropic ↗Reuters ↗

Listen to this edition

Prefer audio? The briefing has chapters and a full transcript.

Follow on Apple Podcasts Follow on Spotify Transcript All episodes →

The briefing

The rest of the morning

4 more stories

Chips and geopoliticsHot

Reuters reports an earlier Huawei chip target as supply runs short

Huawei announced new AI computing systems at its Shanghai conference on September 17. Reuters reports that Huawei now expects its Ascend 960DT chip in the first quarter of 2027, three quarters earlier than planned.

Huawei's rotating chairman Eric Xu told Reuters that the company cannot make enough equipment to meet demand in China. That shortfall limits sales overseas. A larger system on paper does not tell customers whether they can get it, power it, or run their workloads efficiently.

Huawei ↗Reuters ↗

Read article →

Power and policyHot

Australia asks how to keep data-center costs off community bills

Australia announced a public consultation on AI training and large data centers on September 17. It opened at 6 a.m. AEST on September 18 and closes at 5 p.m. AEDT on October 9.

Australia has put these requirements out for consultation. They are not final obligations. The government says they should work alongside state and territory approvals. Communities cannot know how much protection they will get until authorities set the thresholds and enforcement rules.

Australian Department of the Prime Minister and Cabinet ↗Australian Department of the Prime Minister and Cabinet ↗

Read article →

Science and medicineWarm

A new radiology model tackles 146 abdominal scan findings

A study published in Science on September 17 introduces RADAR, an AI model for interpreting abdominal CT scans. The researchers report a mean AUC of 0.913 across 146 findings. The best competing vision-language model in their evaluation scored 0.776.

The results come from the researchers' evaluation. They do not show that routine use improves patient outcomes or that RADAR can replace a radiologist. Others can inspect the published code, but that does not amount to independent clinical validation.

American Association for the Advancement of Science ↗Alibaba DAMO Academy ↗

Read article →

Education and societyWarm

Cornell's classroom pilots can use less AI, too

Cornell announced AI Integration Pilots and an expansion of its AI Undergraduate Fellows program on September 17. A $2 million Dake family gift supports the initiatives, alongside support from the provost's office.

Cornell has announced the classroom trials, but it has not shown that students learn more. Their assignments might look better even if they can no longer explain the work without help.

Cornell Chronicle ↗

Read article →

Analysis

Doing more work is easier to count than doing it well

Anthropic counts research Claude leads. Huawei describes the computing systems it plans to build. RADAR's researchers report how their model scored on scans. None of those measures alone shows whether the work succeeds outside the test.

A delegated task still needs judgment

A research automation percentage describes who does the work. Reviewers still need to check whether the result is correct and worth using.

Computing plans need delivery evidence

A system announcement leaves questions about availability and electricity. Customers and communities need terms they can test against actual operations.

Learning and diagnosis need outcome tests

Students can finish assignments faster without learning more. A scan model can score well without improving treatment. Evaluations need to follow those outcomes.

The watchlist

Signals that could change the read

Independent reviewers examine the research automation measurements.

Customers report deployed capacity and workload performance.

Final rules identify covered facilities and enforceable cost obligations.

Clinical studies and classroom evaluations measure benefits beyond task completion.

Across the desks Who checks the work?

Labs, hospitals and classrooms are testing what they can delegate. Their measures need to show whether the result holds up.

Anthropic research26%Company-reported share led by Claude, with human supervision.

Huawei chip scheduleQ1 2027Planned Ascend 960DT timing, according to Reuters.

Australia consultationOctober 9Public submissions close at 5 p.m. AEDT.

RADAR evaluation146 findingsResearch evaluation, not a patient-outcome trial.

Editorial direction, not a financial index. Each signal is tied to this edition’s reporting.

Edition validated · 5 stories · 9 unique sources

About HotTea & our sources →