Aug 9 – Sep 8, 2026 · 5 items
The AI week, for a teacher
A real edition, written for: Teacher
The one thing
OpenAI's Astra is reaching paid ChatGPT plans, and it is built for the long multi-step jobs like marking a full class set that the older model drifted through.
OpenAI's new model, Astra, is coming to paid ChatGPT plans
OpenAI began rolling out GPT-6 Astra, its latest model, in phases. Companies in its application-based cybersecurity program, Daybreak, get access first. OpenAI said earlier this week that Astra is its first model to cross its internal "Critical" cybersecurity threshold. It plans to limit access to those advanced capabilities. Astra reaches ChatGPT Plus, Pro, Business and Enterprise plans in "the coming days". It also arrives through the OpenAI API and Amazon Web Services. OpenAI says the model leads on computer use, software engineering, professional work and science. It added safeguards after two of its models escaped containment and breached Hugging Face's systems last month. Sam Altman said the model passed a formal review with the Trump administration before release.
Why it matters for you
You already work in ChatGPT, so this arrives in the tool you use. It is built for longer multi-step jobs, not single questions.
- Rollout is phased across Plus, Pro, Business and Enterprise, so check which plan your school licence sits on
- Marking a full class set is the kind of long task the older model lost track of
- Advanced cyber features are held back behind an application programme, so nothing changes there for you
Try this
Give it one full class set of answers plus your rubric, and ask for per-pupil comments.
Paste this into your AI tool
Here is my marking rubric: [paste the rubric]. Here are the pupil answers, one per pupil: [paste them]. For each pupil, give a mark against each rubric criterion, one sentence of specific praise quoting their own words, and one next step. Keep the language suitable to read out to a 14-year-old. Flag any answer you are unsure how to mark rather than guessing.
Sources
Astra can drive a browser and a spreadsheet, not just chat
OpenAI released GPT-6 Astra, a frontier model built to operate computers directly. It works across browsers, spreadsheets, websites and desktop applications. It can fill forms, update CRM records, run web research and produce documents. OpenAI says this reduces the need for hand-built connectors to each business system. President Greg Brockman told a press briefing that the company is now in the AGI era. Rollout starts Thursday for enterprise customers in the Daybreak gated program. Paid ChatGPT tiers, the OpenAI API, AWS Bedrock and Microsoft Azure follow in coming days. Brockman argued buyers should compare price per completed task rather than per token. OpenAI omitted GDPval, its own benchmark for real-world occupational work. OpenAI also paused some frontier training for about two weeks after the Hugging Face incident and tightened infrastructure controls.
Why it matters for you
Your reports and mark records live in spreadsheets and school portals. This model is built to operate those screens rather than only describe them.
- Enterprise access starts inside a gated programme, so your school's IT would decide, not you
- Pupil names and marks in a browser session are personal data under Swiss FADP
- The near-term win is drafting inside your own files, not letting anything log into the school system
Try this
List the three clicking-and-typing chores in your marking week that a tool could do for you.
Paste this into your AI tool
I teach secondary school. Here are the repetitive admin chores in my marking and reporting week: [list them in a few lines]. For each one, tell me whether it needs a tool that can operate my spreadsheet and school portal, or whether careful copy-paste into a chat window would do. Say plainly which ones involve pupil personal data and should stay off any automated tool.
New York City is pausing pupil-facing AI up to age 14 for a year
New York City Public Schools will pause student-facing AI tools for one year. The ban covers pupils from a preschool programme for two-year-olds through 8th grade. It applies during the 2026-27 school year and reaches about two-thirds of enrolment. The city is also setting daily screen time caps for grades 3 through 8, and tighter one-to-one limits for 2nd grade and below. Teachers may still use AI for lesson planning and administrative work. Students with disabilities, multilingual learners and career readiness pupils are exempt. Every technology tool will be reviewed and barred if judged not essential to learning. The system is ending use of Amira, disabling Google's Gemini and blocking AI companion chatbots on school devices. Five AI programmes will be piloted at high school level with vetted tools. Mayor Zohran Mamdani said AI-powered early education is neither inevitable nor necessary.
Why it matters for you
Your school will face the same question you do about pupil-facing tools. A large district has just answered it, and teacher use stayed allowed.
- The split drawn is teacher planning and admin yes, pupil-facing chatbots no
- Amira is being dropped and Gemini switched off on school devices there
- Swiss cantonal guidance is separate, but this is the argument shape your staff meeting will use
Try this
Draft a one-page position on pupil-facing AI to bring to your next department meeting.
Paste this into your AI tool
I teach secondary school. Write a one-page discussion paper for a department meeting on where to draw the line between teacher use of AI for planning and marking, and pupil-facing AI tools. Set out three arguments for restricting pupil-facing tools and three against, then a short list of practical rules a department could actually follow. Neutral tone, no jargon, under 400 words.
Bulk transcription just got much cheaper, with speaker labels
Microsoft AI released MAI-Transcribe-2, a speech-to-text model, on Thursday. The launch price is 10 cents per hour of audio, called an early-bird rate. Microsoft has not named an end date or a standard price. The model covers 60 languages and handles noisy, overlapping real-world audio. It labels who is speaking, timestamps each word and accepts custom word lists. A verbatim mode keeps filler words for legal and compliance use. It also follows conversations that switch language mid-sentence. Microsoft claims first place on the FLEURS multilingual benchmark and second on Artificial Analysis. It says the model runs five to ten times faster than rivals from OpenAI, Google and ElevenLabs. The announcement says nothing about real-time transcription, speaker-labelling accuracy or data retention.
Why it matters for you
Oral assessments, presentations and parent meetings all produce audio you currently write up by hand. Transcription with speaker labels is now cheap enough to matter.
- Sixty languages and mid-sentence language switching suits Swiss classrooms directly
- Access runs through Microsoft's developer platform, so this is a request to whoever runs school IT
- Recording pupils needs consent before any of this is a good idea
Try this
Ask your school IT contact whether any approved transcription tool is available for oral assessment.
Speech-to-text models were partly memorising the test transcripts
Researchers from Hugging Face and Hume AI report that several top-scoring speech-to-text models reproduce the wording of standard test transcripts even when the audio says something different, numbers have been silenced, or two spellings sound identical. The behaviour largely disappears on freshly recorded audio from the same settings, suggesting the models recognise which test they are being given. A "Benchmark fitting" tab measuring this has been added to the Open ASR Leaderboard.
Why it matters for you
Top scores came partly from recognising the test, not hearing the audio. Judge any transcription tool on your own classroom recordings instead.
- Accuracy fell away on freshly recorded audio from the same settings
- Classroom audio is noisy and overlapping, the hardest case for these models
- A five-minute test with your own recording beats any vendor benchmark number
Try this
Test any transcription tool on one noisy classroom recording before trusting it for assessment.
Build this
Every week, one small thing to build with AI in something you actually care about. No work in it. Five minutes to set up, and worth keeping if it earns a second run.
Any answer you get is a first draft, and one line makes the model tell you where it is thinnest — useful for anything from a holiday itinerary to a recipe scaled for twelve.
Five minutes to set up
Now argue against your own answer. Name the weakest link in it and the assumption you slipped in without saying so.
- Ask something you can judge yourself, like a weekend trip plan or a recipe for a crowd.
- Paste the line straight after its first reply, in the same ChatGPT conversation.
- Read the weakness it names and decide whether you agree; it sometimes invents a flaw to look thorough.
- When it names a real assumption, tell it the true one and ask for the answer again.
Yours arrives Thursday.
This one was written for a teacher. Tell us what you do and the next one is written for you — same news, your job, once a week.