AI news6
OpenAI shipped GPT-5.5 "Spud" one week after Anthropic's latest release
OpenAI announced GPT-5.5 on Thursday 23 April, branding it a faster, sharper thinker for fewer tokens than GPT-5.4 and framing it as an agentic model that can autonomously switch between tools across multi-step workflows. It is rolling out to Plus, Pro, Business, and Enterprise subscribers inside ChatGPT and Codex. The release landed exactly one week after Anthropic made Opus 4.7 generally available, and ChatGPT Images 2.0 took the #1 slot across every Image Arena category within 12 hours of its 21 April launch.
techcrunch.com ↗Why it matters
The cadence is the story — frontier labs are now shipping major upgrades weekly, compressing enterprise evaluation cycles from quarters to days.
Anthropic's growing pains mount ahead of OpenAI showdown
Axios reported on Thursday 23 April that Anthropic is facing a cascade of challenges as it heads toward a potential IPO at an $800B valuation, with revenue tripling to $30B this year. A Claude Code software update exposed internal files, the Pro plan briefly removed Claude Code access as a pricing test, and a leaked OpenAI memo from chief revenue officer Tino Cuéllar attacked Anthropic as "elitist" with Sam Altman accusing it of "fear-based marketing." Capacity is straining under demand and the company is simultaneously hiring a six-figure role to negotiate European data center deals.
axios.com ↗Why it matters
The competitive gloves are off between the two frontier labs at precisely the moment Anthropic is trying to project operational maturity to IPO buyers.
Google released new Gemini Enterprise AI agents to challenge OpenAI and Anthropic
Google unveiled a suite of enterprise AI agent tools on Wednesday 22 April designed to help companies automate workflows, in its latest push against OpenAI and Anthropic for the Fortune 500 agent market. The release pairs with Google's new eighth-generation TPU line — split into separate training and inference chips — which it rents to Anthropic, OpenAI, and Meta, and which is engineered for the long context windows and sequential logic that agentic AI requires.
bloomberg.com ↗Why it matters
Google is attacking from two sides at once — infrastructure Anthropic and OpenAI rely on, and the enterprise agent product layer they sell — which raises the stakes on their independent silicon and app strategies.
Anthropic opened a senior Europe role to negotiate gigawatt-scale data center deals
CNBC reported on Thursday 23 April that Anthropic is hiring a six-figure role in Europe to negotiate data center contracts and secure regional compute capacity. The move lines up with the company's just-announced commitment to multiple gigawatts of next-generation Google TPU capacity starting in 2027, built through the Broadcom-Alphabet joint venture. Both Anthropic and OpenAI have publicly accelerated European expansion plans this month.
cnbc.com ↗Why it matters
Sovereign and regional compute is becoming a bottleneck that labs now manage at the contract-negotiation level, not just the engineering level, and Europe is the next front.
OpenAI launched Workspace Agents in ChatGPT for Business, Enterprise, Edu, and Teachers
OpenAI introduced Workspace Agents inside ChatGPT this week, letting teams build and share cloud-based AI agents that handle complex workflows across ChatGPT and Slack with organisation controls, approvals, memory, and analytics. The release is available in research preview on Business, Enterprise, Edu, and Teachers plans. It pairs with the GPT-5.5 release on the agent capability side and tightens OpenAI's move toward "super app" positioning.
releasebot.io ↗Why it matters
Shared cross-team agents with governance built in is the feature enterprise IT has been waiting for before sanctioning AI agents beyond single-user pilots.
European Commission committed €63.2M to AI innovation in health and online safety
The European Commission made €63.2M available on Tuesday 21 April to support AI innovation in health and online safety, alongside its August 2026 AI Act full-applicability deadline. This is paired with standardisation work on GPAI obligations that already came into force August 2025, and sits inside a landscape where California and New York enacted frontier AI laws in late 2025 requiring large developers to publish safety frameworks.
digital-strategy.ec.europa.eu ↗Why it matters
The EU is actively funding applied AI while it finalises the Act's hardest sections, signalling that compliance-ready builders can expect grant capital to de-risk early deployment.
AI in the nonprofit sector4
Humanity AI begins pooled-fund grantmaking in 2026 with $500M from ten foundations
Humanity AI — a $500M five-year initiative co-chaired by the MacArthur Foundation and Omidyar Network — will begin grants from its pooled fund in 2026, with Rockefeller Philanthropy Advisors serving as fiscal sponsor. Founding funders are Doris Duke, Ford, Lumina, Kapor, MacArthur, Mellon, Mozilla, Omidyar, Packard, and Siegel Family Endowment. Grantmaking will prioritise democracy, education, humanities and culture, labour and economy, and security.
macfound.org ↗Why it matters
The single largest philanthropic AI commitment to date is now actually deploying capital, and its five-issue framing is shaping how mid-tier foundations position their own AI grant portfolios.
Hewlett Foundation President called on philanthropy to bridge the AI governance gap
Hewlett Foundation President Amber D. Miller published an op-ed on Wednesday 22 April arguing philanthropy must step up as conversations between Washington and Silicon Valley grow more contentious around AI policy. The op-ed accompanied a grant announcement and positions foundations as the neutral broker between regulators and labs.
aijourn.com ↗Why it matters
This explicitly names philanthropy as a third-party convener in the AI policy stack — a role that had been drifting toward industry self-regulation — and invites more foundations to follow Hewlett's lead.
Nonprofit AI adoption hit 92% but only 7% report major impact, per Virtuous + Fundraising.AI benchmark
The 2026 Nonprofit AI Adoption Report from Virtuous and Fundraising.AI surveyed 346 nonprofits and found 92% using AI tools while only 7% report major organisational capability gains. 81% use AI individually without shared workflows, 79% report small to moderate efficiency gains, and 47% have no AI governance policy. The report frames the gap as an "efficiency plateau" separating experimentation from repeatable impact.
virtuous.org ↗Why it matters
The headline "92% adoption" masks the real sector story — most orgs are stuck in one-off personal prompting, and the path off the plateau runs through documented workflows and governance rather than more tools.
AI for Nonprofits Sprint targets 100,000 nonprofit staff with baseline AI literacy in 2026
The AI for Nonprofits Sprint is working to bring 100,000 nonprofit staff to baseline AI literacy in 2026, with Microsoft, OpenAI, and partner organisations launching free training specifically designed for nonprofit professionals. The programme is explicitly structured as a coalition response to the adoption-vs-impact gap the Virtuous report documented.
nonprofitpro.com ↗Why it matters
A 100K training target coming from the vendor side — paid for by vendors — is the biggest swing yet at the governance-and-workflow gap, but only if completion and post-training impact get measured.