AI news5
Anthropic in talks for $950B funding round, set to surpass OpenAI's valuation
The New York Times reported on Tuesday 12 May that Anthropic is in talks to raise between $30 billion and $50 billion at a valuation as high as $950 billion. That would place Anthropic above OpenAI's most recently reported $825 billion valuation and complete what Sherwood described as a "parabolic" rise from $250B in January 2026. Dario Amodei's company has compounded valuation roughly fourfold in five months.
sherwood.news ↗Why it matters
For the first time since ChatGPT launched, OpenAI is no longer the presumed leader. Enterprise buyers, regulators, and the AI policy world will start treating Anthropic as the default frontier reference, with cascading effects on procurement and standards.
Anthropic surpasses OpenAI in business AI adoption for the first time
Ramp's AI Index data shows Anthropic's enterprise adoption hit 16.7% in February 2026, overtaking OpenAI for the first time since either company existed. Anthropic's share quadrupled over the prior year while OpenAI grew only slightly. Overall business AI adoption hit a record 47.6% in February, with enterprise AI spending doubling to $8.4 billion in six months. Coding tasks account for over half of total enterprise AI spend.
facebook.com ↗Why it matters
Claude Code is the wedge driving Anthropic's adoption lead, which means AI's first defensible enterprise revenue category is being built around Anthropic's tooling rather than OpenAI's. Every AI-buying decision in the next six months will hinge on this data point.
OpenAI ships Codex Windows sandbox, closing the Codex-on-Windows safety gap
On 13 May OpenAI published an engineering deep-dive on building a sandbox to enable Codex on Windows. Previously, Windows Codex users had to choose between approving nearly every command or running Full Access mode with no restrictions. The new sandbox uses an "elevated sandbox" pattern with OS-level isolation, write-only-within-workspace permissions, and no internet access by default. Codex's command-tree-wide enforcement is a meaningful step toward safe agentic coding on enterprise Windows estates.
openai.com ↗Why it matters
Windows is where most enterprise developer machines actually live, and until now Codex was effectively a macOS/Linux product in production environments. This release flips Codex from "developer tool" to "enterprise-deployable AI coding agent" on the world's most common dev OS.
Pentagon expands classified AI work with eight companies, excludes Anthropic
The US Department of Defense (renamed Department of War) signed AI agreements on 13 May with Google, Microsoft, AWS, Nvidia, OpenAI, Reflection AI, SpaceX and one other vendor for use on classified networks. Anthropic was excluded despite being a leading frontier lab, reflecting an ongoing dispute over Anthropic's commercial-use restrictions and refusal to support certain DOD applications. The Marine Corps separately mandated a "Basic AI" training course for all troops the same day.
defensescoop.com ↗Why it matters
The US national-security AI procurement stack is locking in around Anthropic's competitors. For Anthropic, this is the first major commercial cost of its safety-first commercial posture, and for buyers it cleaves the AI vendor map into "Pentagon-approved" and "not".
Marine Corps mandates Basic AI training course for all troops
The US Marine Corps announced on 13 May a mandatory Basic AI training course for all uniformed Marines as part of the Pentagon's pivot to becoming an "AI-first" fighting force. The course covers prompting, prompt-injection awareness, sensitive-data handling, model limitations, and validating model output for operational decisions. Rollout begins immediately and is expected to extend to other branches.
defensescoop.com ↗Why it matters
This is the largest standardised AI-literacy curriculum mandated by any government on the planet to date. The training framework will set the de facto baseline for every other workforce AI literacy programme, including civilian government, healthcare, and the nonprofit sector.
AI in the nonprofit sector5
Paul Ramsay Foundation 2026 Fellowships back five First Nations leaders with up to $250K each
Paul Ramsay Foundation announced its 2026 Fellowship cohort on 12 May. Five leaders driving systemic change to address disadvantage in Australia each receive up to $250,000 over an 18-month fellowship. First Nations leadership is the central thread, with each fellowship explicitly aligned to self-determining First Nations communities. The Fellowship sits alongside PRF's anchor funding role in the NFP Digital Futures launch this week.
nit.com.au ↗Why it matters
This is Australia's largest philanthropic foundation by assets putting its leadership-development money behind First Nations community self-determination during the same week it backed the country's biggest cross-sector NFP digital capability play. The two signals together describe PRF's whole-of-sector thesis for 2026.
Humanity AI announces $18M in new grants to advance people-centered AI
Humanity AI announced $18 million in new grants on 12 May: $8 million awarded to organisations ensuring AI puts people first, and $10 million across a second tranche supporting AI civil-society institutions. This follows Partnership on AI's $500K Humanity AI grant announced the same day. The collaborative philanthropic initiative is now visibly deploying its first $500M wave at speed, focused on civil-society and people-centered AI work.
facebook.com ↗Why it matters
Humanity AI is now the largest single funder of nonprofit-and-civil-society AI work, and its first month of grants signals where the next two years of sector funding will concentrate. Watch the second tranche's recipients closely. They are the AI civil-society shortlist for 2027 institutional growth.
Social Current ships COA Accreditation AI standards for the US social sector
Social Current released its annual COA Accreditation standards update on 12 May, including new standards for responsible and ethical AI use across the US human-services nonprofit sector. The standards mandate a risk-benefit analysis before AI adoption, ongoing stakeholder engagement (including impacted communities), human oversight for high-stakes decisions, and documented AI guidance with staff training. The framework is anchored on four principles: stakeholder engagement, transparency, human oversight and accountability, and risk and liability exposure.
social-current.org ↗Why it matters
COA is the largest accreditor of US human-services nonprofits. Embedding AI standards in COA's accreditation framework operationalises responsible AI across thousands of human-services organisations in a way no voluntary toolkit ever did.
AWS Imagine Nonprofit London pivots messaging to "Build responsibly: governance, ethics, accountability"
At AWS Imagine Non-Profit London on 12 May, AWS's core sector message centred on "Build responsibly with AWS: Governance, Ethics, Accountability" rather than the agents-and-models pitch that anchored prior years. The conference featured case studies from across AWS's nonprofit customer network framed around responsible deployment, with the AI for Nonprofits Network's live coverage noting the shift as a material change in tone.
aifornonprofitsnetwork.substack.com ↗Why it matters
The hyperscaler-led nonprofit-AI conversation has visibly moved from "look what's possible" to "deploy safely." For NFP operators, this signals that AWS-funded grant-credit programs and Imagine grants will increasingly hinge on governance maturity, not just creativity.
Vu Le calls for serious AI conversations in nonprofit and philanthropic sector
Nonprofit-sector commentator Vu Le published a NonprofitAF essay on 12 May arguing the sector needs much more serious conversations about AI and philanthropy. Le argues current sector AI discourse is dominated by either uncritical tool celebration or vague ethical hand-wringing, with too little attention to power, labour, and how AI reshapes who decides what gets funded. Le is the most-quoted sector voice on philanthropy critique in North America.
nonprofitaf.com ↗Why it matters
Sector influencers are starting to push back on AI vendor framings dominating the conversation. Expect more critical commentary like this through Q3 2026 as nonprofit AI moves from novelty to standard practice and the labour and equity tradeoffs become harder to defer.