AI news6
Stripe tests coding agents on 69 real payment scenarios
Stripe's 19 November evaluation tested 14 model-and-agent combinations on 69 payments, subscriptions and marketplace scenarios. Its average score was 78.1%, but requiring every necessary criterion to pass reduced the average to 29.4%. All 14 setups improved with Stripe's agent plugin; these are Stripe's own benchmark results.
stripe.dev ↗Why it matters
A checkout that runs can still miss payment failures or subscription changes, making domain-specific checks more useful than a generic coding score.
Gemini adds on-demand tool loading in its Interactions API
Google's 19 November Tool Retrieval preview loads detailed tool schemas only when an agent needs them, while keeping tool names and descriptions available. It supports Gemini 3.6 and later models, including Gemini 4, and remote MCP servers. Standard token charges still apply; Google gives no measured savings figure in the announcement.
blog.google ↗Why it matters
Large tool libraries can consume less prompt space, but teams still need to measure the savings and check which actions their agents can execute.
Ideogram 4.6 adds precise image editing at up to 4K
Ideogram released version 4.6 on 19 November with Generate and Precise Edit API endpoints. Generate accepts up to five reference images, and Precise Edit can use a mask to target a change. Output supports up to 4K, while the Very low quality setting stops at 2K.
ideogram.ai ↗Why it matters
Repeated edits offer a practical way to refine campaign visuals while checking that faces, products and surrounding details stay consistent.
Mathematician publishes a claimed Claude-assisted Riemann proof
Levent Alpöge's 20 November announcement says he and Claude found a proof of the Riemann Hypothesis and formalised the argument in Lean. His four-page note links the code and describes the argument as exceptionally long. This is a claimed result awaiting independent checking, not an established resolution of the problem.
alpo.ge ↗Why it matters
AI-assisted mathematics is advancing, but a formal proof claim still needs scrutiny of the statement, dependencies and verification process.
Archer raises US$10 million for an AI agent serving both sides of hiring
Archer, from hackajob, announced US$10 million led by Volition Capital with AVP on 20 November. The company reports more than 700,000 candidate sign-ups over 12 months and 10,000 introductions a week, with hiring decisions staying with recruiters. A larger candidate launch is planned for January, and general release of Archer OS for hiring teams is planned for the first quarter of 2027.
hackajob.com ↗Why it matters
Recruiting agents are moving into sourcing and qualification, where useful introductions and human hiring decisions matter more than application volume.
ChatGPT Space begins rolling out globally to adult Plus users
OpenAI's 20 November release notes expand Space to Plus users aged 18 and over worldwide. It supports interactive pages and presentations that can update from connected apps, with collaboration for other people who have Space access. Shared work requires permission and a signed-in account with Space access; Free users cannot open it, and public links are unavailable.
help.openai.com ↗Why it matters
Living project documents become accessible to more subscribers, but sharing still depends on each collaborator's account access.
AI in the nonprofit sector3
Tarbell lists 22 November as its AI journalism fellowship deadline
Tarbell's 2027–28 fellowship page lists 22 November 2026 as the application deadline, without stating a closing clock or timezone on that page. The full-time programme combines training from June 2027 with a nine-month newsroom placement; standard salaries range from US$75,000 to US$120,000. It primarily targets applicants with work rights in the US, UK or Hong Kong, and visa sponsorship is uncommon.
tarbellcenter.org ↗Why it matters
This is a closing-date reminder of the 9 November item, so interested applicants should confirm the cutoff in the official form rather than assume Sydney midnight.
HLA offers 20 funded NetHope places with access to AI training
The Humanitarian Leadership Academy is offering up to 20 funded, 12-month NetHope Connected memberships, with applications due at 10:59 am AEDT on Thursday 26 November. Eligible nonprofits generally have 5–200 staff, annual income below US$2 million and basic digital capacity; local and national Global South organisations receive priority. Members become eligible for free access to a 90-day AI for Nonprofits Accelerator; membership is not itself a cash grant.
humanitarianleadershipacademy.org ↗Why it matters
The call offers smaller organisations a supported route into AI skills and digital capacity, with an applicant-session recording from 18 November available to clarify fit.
Sanofi offers €50,000–100,000 a year for youth climate-health projects
Sanofi Foundation's call, opened on 10 November, funds projects for 12–24 months and explicitly includes youth-led approaches using AI and community data. Applicants must be based in Chad, Ethiopia, Kenya, Malawi, Senegal, Uganda or Zimbabwe, and at least 80% of the budget must support local implementation. Concept notes close at 9 am AEDT on Saturday 12 December, equivalent to 11 December at 11 pm CET.
foundation-sanofi.com ↗Why it matters
This is a targeted climate-health opportunity for eligible local organisations using AI, with the AI eligibility and award terms confirmed in the funder's guidelines.

