AI news5
Anthropic livestreams Code with Claude London on Tuesday, ahead of expected product moves
Anthropic confirmed its Code with Claude London developer event runs Tuesday 19 May at the Old Truman Brewery, with a free livestream open to anyone who registers. The day is structured around three tracks (Research, Claude Platform, Claude Code), with an Extended day on 20 May for independent developers and early-stage founders. London is the second stop in the 2026 series after San Francisco on 6 May, with Tokyo to follow on 10 June.
claude.com ↗Why it matters
The SF event two weeks ago dropped Claude Managed Agents and the SpaceX Colossus compute deal, so Tuesday is the next likely vector for new product or pricing moves that ripple straight into nonprofit tooling stacks.
Google previews Gemini-everywhere stack 48 hours out from I/O
The Android Show I/O Edition aired over the weekend, with Google branding the AI layer "Gemini Intelligence" and confirming it will run beneath Android, Wear OS, Android Auto, and Android XR starting on Galaxy and Pixel hardware this summer. Google also announced Googlebook, a premium AI-PC line shipping in autumn with Magic Pointer and Create My Widget features baked into the OS. Full I/O keynote opens Tuesday 19 May, 1pm ET.
blog.google ↗Why it matters
The naming shift from "Gemini app" to "Gemini Intelligence" signals Google reframing Gemini as infrastructure rather than a product, which changes the comparison set when nonprofits weigh Workspace AI against Claude or Copilot.
OpenAI ships Codex CLI update with modal Vim editing, fork picker, and richer status line
Releasebot logged a major Codex CLI update on Saturday 16 May, adding modal Vim editing, a redesigned resume and fork picker, and richer status-line controls. The release lands while OpenAI consolidates ChatGPT, Codex, and the developer API under one product team led by Thibault Sottiaux, aiming at a single super-app with Atlas browser integration.
releasebot.io ↗Why it matters
Codex-as-default for in-terminal agentic work is creeping closer to Claude Code parity, and the Vim modal edit lowers friction for ops-style users. Anyone running both side by side should re-benchmark this week.
EU AI Act Digital Omnibus deal moves toward formal adoption, pushing high-risk deadline to 2027
The European Council and Parliament's provisional Digital Omnibus agreement from 7 May is now moving toward formal adoption, delaying high-risk AI rules until 2027 and pushing national regulatory sandbox deadlines out to 2 August 2027. The package adds new prohibitions on AI systems that generate non-consensual intimate imagery and child sexual abuse material, including nudifier apps. Texts are now in legal-linguistic revision.
consilium.europa.eu ↗Why it matters
The compliance breathing room buys time for nonprofits deploying high-risk-classified tools (employment screening, education ranking) but does nothing for the nudifier ban, which lands as soon as adoption clears.
Mary Meeker drops 340-page BOND AI Trends report with Stripe fraud-catch jump cited
Mary Meeker's first major trends report in five years notes ChatGPT hit 800 million users in 17 months, training costs reaching $1 billion per frontier model, and inference costs dropping 99 percent per million tokens over two years. Stripe's fraud-detection assistant jumped from 57 percent catch rate to 97 percent after AI integration, per the report. Open versus closed source is now framed as "freedom vs control, speed vs safety, openness vs optimisation."
benchmarkit.ai ↗Why it matters
The 99 percent inference-cost drop is the number that matters for nonprofit budget planning, and the Stripe example is a clean small-team productivity story.
AI in the nonprofit sector4
UNICEF expands Tinkering with Tech AI-skills programme via Arm and Raspberry Pi as decade-long AI in Play platform
UNICEF's Helsinki-based Global Learning Innovation Hub is scaling Tinkering with Tech into new countries with a stronger AI-skills focus, in partnership with Arm and Raspberry Pi under a decade-long platform called AI in Play. The expansion shifts from pilot delivery to broader system integration, with Arm playing a strategic role.
edtechinnovationhub.com ↗Why it matters
This is one of the few large-scale AI-literacy programmes built around bottom-up community tinkering rather than top-down curriculum, which gives nonprofits a concrete counter-example when asking what grassroots AI training actually looks like.
UNICEF and Korea move toward letter of intent on Global AI Hub for humanitarian challenges
UNICEF's Executive Director visited Seoul in late April to formalise participation in Korea's Global AI Hub, which seeks cross-border cooperation on AI solutions for humanitarian challenges, with both sides moving toward signing a letter of intent. UNICEF framed this as a significant commitment to shaping the initiative.
koreatimes.co.kr ↗Why it matters
Korea is positioning as a third pole in AI-for-humanitarian work alongside US and EU efforts. An LOI with UNICEF is the signal Australian aid orgs should watch when planning multilateral funding pitches.
Skoll Foundation names ChildLife Foundation, SmartStart and Indus Action 2026 Skoll Award winners with $1.5M each
The Skoll Foundation announced the three winners of the 2026 Skoll Award for Social Innovation: Pakistan's ChildLife Foundation (paediatric emergency telemedicine), South Africa's SmartStart (early childhood development franchise), and India's Indus Action (welfare access for low-income families). Each receives $1.5 million in core support.
prnewswire.com ↗Why it matters
The selection skews hard toward systems-change orgs with AI-enabled scale models rather than direct-service grantees, signalling where Skoll wants its capital pointed for the next cycle.
GiveDirectly uses Google flood-prediction AI to ship anticipatory cash before crises
GiveDirectly is now using Google's AI-based flood prediction tool to disburse cash payments before humanitarian crises hit rather than after, in a working anticipatory-action model. Devex framed this as a practical, deployed example of AI cutting the response-time gap in humanitarian cash transfer.
devex.com ↗Why it matters
This is the cleanest "Google.org AI tool actually working in a deployed nonprofit context" story of the week and gives a concrete pre and post pattern to cite when nonprofits ask whether weather AI is real or hype.