Back to School
Credentials Nobody Approved
6 items · 17–24 August 2026 · primary sources onlyTwo sets of credentials were issued this week and no institution voted on either. Students got theirs from consumer products: Google made its $19.99-a-month AI Pro tier free for twelve months to US undergraduates, and OpenAI began routing anyone it estimates to be 13–17 into a separate teen product — the same day it put advertising into the free tier across 31 European markets. Agents got theirs from admin consoles: Workspace Studio gave automated flows their own identities, and AWS gave agents a wallet. The European rules governing the education and employment side of this bind from 2 December 2027, so for the next fifteen months the defaults are set by product decisions — and institutional policy will be whatever those defaults turn out to be.
Google made its paid AI tier free to US college students for a year, with a card on file
Eligible US college students get twelve months of Google AI Pro (normally $19.99/month, with 5 TB of storage); students in 140+ other markets get the lower AI Plus tier. Shipped, not preview. A payment method is required at sign-up, redemption closes 31 December 2026, and the plan auto-charges standard rates after the year.
This settles the tool question for a full academic year before any curriculum committee meets. By the time an institution finishes evaluating an enterprise agreement, its students will have a year of habit and 5 TB of coursework in a personal account it does not administer and cannot audit. The card-on-file detail matters more than the price — the cohort converts mid-degree in August 2027, and the students least able to absorb $19.99 a month are the ones who will have built the most dependence.
Whether universities answer with enterprise agreements, or let the consumer tier become the de facto standard.
OpenAI split teenagers into their own ChatGPT, and left schools on a separate product
Users who report being 13–17, or whom OpenAI's age estimation places under 18, are routed automatically into ChatGPT for Teens: Study Mode, quizzes, Study Hours, and safeguards that avoid romantic language or implying feelings. Linked parents can set Quiet Hours and receive high-risk safety notifications. Schools stay on the separate, institution-managed ChatGPT for Teachers.
The routing is automatic and inferred, which makes it age assurance operating on students without an institutional decision behind it — and it will misclassify in both directions, most consequentially for 18-year-old first-years. The structural point is the split: the teen experience has a parent as administrator, the school product has an IT department. A student now has two differently-governed ChatGPTs, and coursework will move between them.
What happens to a teen account's history at 18, and whether the age estimate is appealable.
ChatGPT advertising went live across 31 European markets, on the free tier
Ads expanded to 31 European markets, with geo-targeting and custom audiences. They appear for Free and Go users; Plus, Pro and Enterprise stay ad-free. OpenAI states conversations remain private from advertisers and that it does not sell customer data.
The free tier is the student tier, and it is now the advertising tier — announced the same day as the teen product, which the post does not say is exempt. This puts a commercial-content surface inside the tool many institutions have been recommending because it costs nothing, in the jurisdiction where deployers have owed AI transparency duties since 2 August 2026. Recommending the free tier is now recommending an ad-supported one.
An explicit statement on whether teen and education accounts see ads.
Google Workspace gave agents their own identities, and admins a revocation dashboard
Workspace Studio flows now run under their own least-privilege agent identity instead of inheriting the owner's full rights. Admins get a dashboard to suspend flows or revoke individual OAuth scopes, flow-context audit logging, enforced human-in-the-loop confirmation, and DLP. Admin settings rolled out 17 August; end-user features 20 August (Rapid Release) and 1 September (Scheduled). Included in Education editions.
This is the governance layer arriving after the capability — the usual order, and the reason so many institutions have automations still running under a departed employee's account. Least-privilege identity plus scope revocation turns an agent into an auditable actor rather than a person's delegated shadow, which is what makes a real policy about them writable. It also means someone has to own the register of who runs what.
Whether IT teams inventory existing Studio flows before the 1 September Scheduled Release date.
AWS put agents on a payment rail, generally available
Amazon Bedrock AgentCore Payments reached GA, letting agents autonomously pay for APIs, MCP servers and paywalled content. Users grant explicit spending delegation, the agent only ever holds short-lived tokens, and each payment session carries a maximum spend and an expiry with audit trails.
Delegated spending authority is where agent governance stops being an IT question and becomes a finance and audit one. The dynamic per-consumption pricing this enables is the quieter signal: it erodes per-seat licensing, and procurement and grant-compliance processes are built on per-seat assumptions.
The first institutional finance policy that names an agent as a spender.
OpenAI paused frontier training after a model hit its critical cyber threshold
OpenAI disclosed that its “Astra” model was determined on 7 August to meet the Critical cybersecurity capability threshold, and that it has slowed scaling — a two-week pause in reinforcement-learning training, with its largest planned RL run still on hold — plus new isolation and 30-minute alerting requirements.
A lab pausing its own largest training run is the clearest evidence yet that the binding constraints on frontier development are currently self-declared. That is the same gap an institution hits when it asks a vendor which safety commitments actually oblige them.
Whether the paused run resumes without a public statement.
- OpenAI · 19 Aug — Zero Data Retention continues, and Private Safety Processing keeps flagged content under customer keys with no OpenAI staff access. White paper due September; relevant to IRB-governed work. source
- Google for Education · 19 Aug — Study notebooks driven by diagnostic quizzes, and Classroom controls to tailor student AI to the curriculum. Search study tools announced, not shipped. source
- OpenAI × CodeAI · 18 Aug — A free year-long high school AI course, an Hour of AI aimed at millions of students, and a youth advisory council. source
- Slack and GitHub · 20–21 Aug — Slack Code channels GA for Claude, Devin, Copilot and Vercel; Copilot in Slack and Teams in preview, opening pull requests under an app identity. source
- Salesforce · 19 Aug — Data 360 MCP Server GA with roughly 200 APIs, giving authorised agents write access to systems of record. source
- arXiv · 18 Aug — A twenty-jurisdiction map of AI governance finds roughly one provision in five sits in binding law, and that outside the EU no jurisdiction mandates developer evaluation duties. source
The EU AI Act's high-risk regime covering education and employment applies from 2 December 2027, deferred from August 2026 by the AI Omnibus. Article 50 transparency duties have bound deployers since 2 August 2026. Commentary still asserting an August 2026 incident-reporting duty for education AI is stale. Commission
Generated 24 August 2026 · Window: 17–24 August 2026
Six scored items, six also-noted · Primary sources only