Aug 9 – Sep 8, 2026 · 6 items
The AI week, for a lawyer
A real edition, written for: Lawyer
The one thing
EU AI Act transparency duties are now enforceable, so any client of yours selling into the EU needs those disclosure obligations reflected in the contracts you are drafting this month.
OpenAI's GPT-6 Astra works inside documents and spreadsheets, 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
You draft and review contracts in ChatGPT today. This model is built to work through documents end to end, not answer questions about them.
- Rollout reaches paid ChatGPT tiers over the coming days, so no admin request needed on the individual plan
- Enterprise access starts through a gated programme, so your firm's IT may see it later than you do
- Client confidentiality still decides what you can paste, whatever the model can do
Try this
Give it a full markup round on a live NDA and check every changed clause yourself.
Paste this into your AI tool
Here is a draft NDA sent to us by the other side: [paste it]. Here are our standard positions: [paste your firm's key requirements, or list them briefly]. Produce a clause-by-clause table: clause reference, what it currently says, why it departs from our position, and suggested replacement wording. Flag anything ambiguous rather than rewriting it silently. Note anything governed by Swiss law that needs a second look.
Microsoft's new transcription model prices bulk audio at a tenth of what it was
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
Negotiation calls and client meetings are your record-keeping problem. Bulk transcription with speaker labels and timestamps just became cheap enough to do routinely.
- Speaker labelling and word timestamps are what make a call transcript usable as a file note
- A verbatim mode keeps filler words, which matters when the exact wording of a concession is disputed
- Microsoft says nothing about data retention, so client audio needs that answered before anything is uploaded
Try this
Ask your IT or knowledge lead whether firm-approved transcription now covers client calls, and on what retention terms.
EU AI Act transparency duties are now enforceable, and Switzerland is outside them
The European Union's AI Act moved into enforcement on 2 August 2026, with disclosure duties now applying to chatbots and to content produced by AI. The EU's AI Office can request information from covered companies and ask for access to their models, though it has not yet pursued anyone for misconduct. Anthropic, Google, Meta, OpenAI and Microsoft have each described compliance steps, including watermarking generated text. Rules for high-risk uses such as education, biometrics and migration arrive only in December 2027 and August 2028.
Why it matters for you
Your clients selling into the EU now face live disclosure duties on chatbots and AI-generated content. Swiss-domiciled clients are not exempt if they sell in.
- The AI Office can now formally request information and model access from covered companies
- High-risk rules land in December 2027 and August 2028, so contracts signed now should anticipate them
- This is a reason to add AI-disclosure and cooperation warranties to your supply and services templates
Try this
Draft an AI-transparency warranty and cooperation clause for clients selling into the EU.
Paste this into your AI tool
Draft two clauses for a Swiss-law commercial services agreement where the supplier uses AI systems and the customer sells into the EU. First: a warranty that the supplier complies with EU AI Act transparency and disclosure duties, including labelling AI-generated content and disclosing chatbot interactions. Second: an obligation to cooperate with regulator information requests and pass through required documentation. Keep the drafting tight and neutral. Then list, in plain terms, the three points a client is most likely to push back on.
Sources
Compliance consultants publish an AI checklist for regulated advisors
US investment advisors face SEC examinations that treat artificial intelligence as a top focus this year, even though no rule written specifically for AI exists after the regulator dropped its proposal. Compliance consultants recommend written rules on human review of AI output, tighter terms for outside service providers, accurate marketing claims and a named owner for AI use. A survey of compliance officers found fewer than half of firms require a person to check AI results, and it remains unsettled whether AI-generated client meeting notes count as official records.
Why it matters for you
Your financial-services clients are being told to document human review, vendor AI terms and named ownership. That is contract work, and it is yours.
- Under half of surveyed firms require a person to check AI output, so vendor terms are doing the work instead
- Whether AI meeting notes count as official records is unsettled, which affects retention clauses you draft
- The US position is not binding in Switzerland, but clients with US arms will ask you to mirror it
Try this
Build a reusable AI-clause checklist for vendor contracts your regulated clients sign.
Paste this into your AI tool
I am a commercial lawyer advising clients who buy AI tools from third-party vendors. Produce a review checklist for the AI-specific provisions in a vendor contract, covering: human review of outputs, training on customer data, sub-processors, retention of prompts and outputs, accuracy and marketing claims, audit rights, and allocation of liability for AI errors. For each item give the question to ask and the fallback position to accept. Keep it to one page and write it so I can use it on a call.
One reusable prompt got harmful answers out of most major models
An independent researcher turned a safety research prompt into a jailbreak that works across many models. Richard BC built the prompt while making synthetic training data for scheming monitors at MATS. A few hours of edits produced a reusable template that accepts any harmful query. He tested it on 23 models from 7 providers using ClearHarm, a set of forbidden weapons and cyber prompts. Nearly every model produced at least one fully harmful answer. Newer Anthropic models and Meta's Muse Spark 1.1 refused throughout. Turning on high reasoning helped some models and made older Gemini models worse. Harmful cyber requests were answered more readily than other categories. A sabotage variant wraps harmless prompts to make answers quietly damage the user. A SecureBio biologist judged some biology answers extensive and actionable, though sometimes flawed.
Why it matters for you
You paste client material into ChatGPT. This research shows the guardrails on these tools are weaker and more uneven than vendors imply.
- A sabotage variant wraps a harmless request so the answer quietly damages the user, which is a review problem not a refusal problem
- It is direct support for a firm rule that AI output is always read by a person before it leaves the building
- Useful ammunition if you are asked to sign off a vendor's safety claims in a contract
Try this
Write a two-paragraph note to your practice group on why AI drafts get a human read before sending.
Paste this into your AI tool
Write a two-paragraph internal note for lawyers in a mid-size commercial firm. Point: AI drafting output must be read line by line by the responsible lawyer before it goes to a client or counterparty, because model safety and accuracy controls are inconsistent and failures can be subtle rather than obvious refusals. Plain professional tone, no hype, no bullet points. End with one sentence on what to do when a draft looks right but the lawyer cannot verify a cited point.
Nvidia is buying Hugging Face for just under $13bn, closing in 2027
Nvidia has agreed to buy Hugging Face, a hosting hub for open AI models and datasets. Nvidia confirmed the definitive agreement on Thursday after weeks of industry speculation. The deal is expected to close in the first half of 2027, subject to regulatory approval. Nvidia pledged to keep the platform open to all model makers, developers and users. Jensen Huang said open models let organisations build without training everything themselves. Hugging Face CEO Clement Delangue said the platform needs more compute, support and visibility. Forrester analyst Charlie Dai expects openness to be preserved at first. He advised enterprise users to watch for later shifts and to assess the risk of deeper ties to Nvidia's own tooling. Critics also question the circular flow of money between AI suppliers and their customers.
Why it matters for you
Your clients' AI vendors often build on open models hosted there. A change of ownership is a change-of-control and continuity question in their contracts.
- Close is expected in the first half of 2027 subject to regulatory approval, so there is time to paper around it
- Analysts advise watching for later shifts in openness, which is exactly what a continuity clause is for
- Worth a line in vendor diligence: which model source does this product depend on
Try this
Check whether your standard change-of-control clause covers a supplier's upstream model provider.
Paste this into your AI tool
Review this change-of-control clause from a Swiss-law software supply agreement: [paste the clause]. Assess whether it protects the customer when a supplier's upstream AI model host or model provider is acquired, rather than the supplier itself. Identify the gap in plain terms, then propose replacement wording that captures material changes in the supplier's key AI dependencies. Keep it short and note any drafting risk in making it too broad.
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.
A drinking order for the bottles already in your rack: what to open this month, what to leave, and why.
Five minutes to set up
Here is what is currently in my wine rack or cupboard: [list each bottle: producer, region, grape if you know it, vintage]. Here is how I drink: [rough note, e.g. two weeknights a week, occasional dinner for six]. Using only these bottles, build a drinking order for the next three months. For each bottle give a month, a reason tied to its likely drinking window, and one simple food pairing I could actually cook. Name the bottles that should wait longer and say what they are waiting for. Do not suggest anything I do not already own.
- Write the list straight off the rack, including the bottles you have forgotten why you bought.
- Run it in ChatGPT, which can hold the whole list and re-order it when you push back.
- Check any drinking-window claim on a bottle you care about before you follow it — this is where it will be confidently wrong.
- When the order feels wrong, tell it which bottle you actually want open first and make it rebuild around that.
Yours arrives Thursday.
This one was written for a lawyer. Tell us what you do and the next one is written for you — same news, your job, once a week.