AI news5
Linux Foundation launches Agent Name Service to give AI agents trusted identities on the internet
The Linux Foundation announced on June 23 its intent to launch the Agent Name Service (ANS), an open standard that provides trusted identity, verification, and discovery for AI agents operating across the internet. ANS is built on DNS, the same globally distributed infrastructure that already processes more than 100 million queries per second, and extends it with a verification layer that lets systems confirm who an agent represents, what permissions it holds, and whether its code and operational history remain unchanged. The framework supports decentralised identifiers (DIDs) and Legal Entity Identifiers (LEIs), and the project is open to enterprises, AI developers, infrastructure providers, and security researchers. A World Economic Forum dataset cited in the announcement found that 82% of executives plan to adopt AI agents within three years, yet no agreed identity standard currently exists for them.
devopsdigest.com ↗Why it matters
agentic AI is moving into production in every sector, including nonprofits using Claude Code, Cursor, and similar tools for automation, yet there is currently no standard way for systems to verify whether an AI agent is authorised, authentic, or tamper-free. ANS fills that gap using infrastructure organisations already rely on, which is why the project attracted GoDaddy as a named co-developer rather than requiring a new lookup network from scratch.
Meta is the sole holdout as Trump's voluntary AI model security review framework goes live
The Trump administration is pressing Meta to submit its AI models for voluntary 30-day pre-deployment review by federal security agencies, with Meta now the only major AI lab that has not agreed to the process, according to reporting by the New York Times and Reuters on June 23. The framework was established by Trump's June 2 executive order on AI innovation and security, which originally proposed a 90-day review window; that was cut to 30 days after sustained lobbying by Elon Musk and Mark Zuckerberg. OpenAI is already in active review, with GPT-5.5 and GPT-5.5-Cyber being evaluated by the Office of the National Cyber Director and the Office of Science and Technology Policy. The order explicitly does not create mandatory approval requirements and does not restore the Biden-era pre-deployment safety testing mandates that Trump revoked on day one of his second term.
reuters.com ↗Why it matters
Meta being the only holdout creates an asymmetric reputational exposure that its rivals do not share. If the voluntary framework is later strengthened into mandatory requirements, Meta's prior refusal will be cited as evidence of bad-faith resistance, while OpenAI, Google, Anthropic, and Microsoft can point to their cooperation. For nonprofits that rely on Meta AI products, Meta's opt-out means those products will receive less independent federal security scrutiny than competitors.
Google releases Diffusion Gemma: an on-device AI model that generates text four times faster by processing chunks in parallel
Google released Diffusion Gemma alongside the Gemini 3.5 family as an experimental on-device text generation model. Unlike standard autoregressive models that generate one token at a time from left to right, Diffusion Gemma processes 256-token chunks simultaneously, achieving up to four times the generation speed at a trade-off in output coherence. The model runs locally without API calls and includes inline editing and code infilling features aimed at developer tooling. Diffusion Gemma is not a replacement for cloud-based Gemini models and is explicitly positioned for tasks where speed matters more than precision: drafting, autocomplete, rapid code comment iteration, and similar high-frequency low-stakes outputs.
buildfastwithai.com ↗Why it matters
a model designed for fully local, no-API-call deployment removes a structural barrier for nonprofits handling sensitive data, including healthcare, legal aid, and child welfare organisations that cannot route content through external servers. The four-times speed gain matters in contexts like intake document summarisation or screening tools where generation latency currently limits real-world adoption.
Gemini 3.5 Pro runs past Google's June launch target as the company's AI narrative faces pressure
Gemini 3.5 Pro, which Google CEO Sundar Pichai explicitly committed to a June 2026 launch at Google I/O on May 19, remains in limited Vertex AI enterprise preview as of June 24 with no general availability confirmed. Confirmed specifications include a two-million-token context window, a Deep Think reasoning mode gated to Ultra subscribers at $250 per month, and pricing of $15 per million input tokens and $60 per million output tokens. Prediction markets place a 50 to 55% probability on a GA launch before June 30. The potential delay lands as Google has already lost AlphaFold Nobel laureate John Jumper to Anthropic and Transformer architecture co-author Noam Shazeer to OpenAI in the same week, making Gemini 3.5 Pro the company's most important near-term narrative asset.
buildfastwithai.com ↗Why it matters
a July slip would mean Google's AI story in June consists of two headline researcher departures, a delayed flagship model, and a competitor about to launch GPT-5.6. For organisations evaluating model selection, Gemini 3.5 Pro's two-million-token context window is a genuinely differentiating feature for long-document workflows, and clarity on the launch date matters for procurement planning.
GPT-5.6 expected by end of June with a 1.5M token context window and pricing below Anthropic
Strong pre-release signals point to an OpenAI GPT-5.6 launch before July 1, with leaked specifications indicating a 1.5 million token context window, faster Codex responses, and pricing set to undercut Anthropic's equivalent tier. OpenAI has not officially confirmed the model or launch date as of June 24; the strongest signals come from API spec leaks reported by AI monitoring publications and a LinkedIn post from AI Insiders citing the company's intent to position GPT-5.6 below Anthropic on price while above GPT-5.5 on capability. GPT-5.5-Cyber, a security-focused variant already in federal review under Trump's June 2 EO, suggests the GPT-5.6 family will include specialised downstream variants. If it ships before July, GPT-5.6 would arrive the same week Noam Shazeer, co-author of the Transformer architecture, starts his role as OpenAI's Lead for Architecture Research.
aitoolsreview.co.uk ↗Why it matters
a 1.5 million token context window at below-Anthropic pricing directly changes the economics of long-document processing for nonprofits. Organisations currently making model-selection decisions based on cost per query and context length should wait for the launch before committing, as the pricing gap between providers is expected to narrow materially this week.
AI in the nonprofit sector4
Blackbaud Institute: only 10% of nonprofits are "AI-Adaptive" as a four-gap effectiveness crisis widens across the sector
The Blackbaud Institute released a new report on June 24 drawing on surveys of 1,389 social impact professionals and donors conducted in March 2026. While 85% of social impact professionals report using AI at work, only 33% believe their organisation is using it effectively. The report identifies four structural gaps that separate organisations generating real outcomes from those stuck in individual experimentation: an effectiveness gap (AI use doesn't translate to organisational results), an infrastructure gap (25% of organisations rely exclusively on free AI tools), a data readiness gap (fewer than 20% rate their organisation's data health as excellent), and a transparency gap (76% of donors expect organisations to disclose when AI is used, but only 26% of organisations currently do so). The 10% of organisations classified as AI-Adaptive, those with systemic and governed AI use, save $621 per employee per week on average versus $500 for the field overall and reinvest those savings into mission delivery and revenue growth. Blackbaud also launched an AI Coalition for Social Impact and a free AI certification programme for sector professionals.
prnewswire.com ↗Why it matters
the 85% adoption versus 33% effectiveness gap is the most precise measurement yet of where the nonprofit sector is stuck. The transparency finding is particularly sharp: donors already expect disclosure and the sector is not providing it, which creates a quiet trust exposure at exactly the moment AI use is becoming visible through automated donor outreach, AI-generated communications, and agentic workflow tools.
Eight years after Schwarzman's $350M gift, MIT is asking whether philanthropy can bend AI's trajectory
In 2018 Stephen Schwarzman gave MIT $350 million as the lead gift toward a $1 billion total that launched the Schwarzman College of Computing, with an explicit goal of producing ethically grounded AI leaders who would shape how the technology is developed and deployed. The Chronicle of Philanthropy published a feature on June 24 evaluating whether the investment has worked. The college has created campuswide interdisciplinary programs integrating AI ethics, social implications, and policy into computer science curricula, and the $1 billion goal is now fully raised. Devin Kim, formerly of Elon Musk's xAI and now president of the Center for AI Safety, told the Chronicle that academia's core value is independent research credibility: "The number one thing academia does is it establishes independent research credibility outside of what the companies are telling you." The article notes that nearly 60% of Americans believe AI poses high risk to society, a figure that has risen over the eight years the college has been building its programs.
philanthropy.com ↗Why it matters
the question of whether a single large philanthropic gift can change the direction of a technology sector is directly relevant to every foundation currently weighing where to put AI-related resources. The MIT story suggests the answer is nuanced: a $1 billion institutional investment can build lasting structural capacity for independent research and interdisciplinary leadership development, but it operates on a decade-scale timeline while commercial AI moves in 18-month cycles.
AI for Good Global Summit opens in 12 days: 300 speakers, Innovation Factory finals, and a first-ever AI youth zone
The ITU AI for Good Global Summit runs July 7 to 10 at Palexpo Geneva, with 300 speakers from government, industry, academia, and civil society expected, plus an Innovation Factory finals round in which national competition winners pitch AI-for-good solutions on the global stage. Australia's Enterprise Monkey, which won the national competition in Perth on May 27 with its Agents for Humanity platform, will represent Australia at the finals alongside winners from other participating countries. A first-ever Youth Zone will run 50-plus hands-on AI workshops for participants aged 13 to 17. Online attendance is available at 10 CHF per session (approximately $11 AUD), making the event accessible to practitioners who cannot travel to Geneva. The Summit is organised by the International Telecommunication Union, the United Nations' specialist agency for digital technologies.
aiforgood.itu.int ↗Why it matters
the AI for Good Summit is the closest thing the sector has to a global convening on AI for social benefit, and the Innovation Factory model, in which national winners advance to a Geneva finals round, is creating a pipeline of vetted AI-for-good startups across multiple countries. With 12 days to the opening, registration is still open for both in-person and online attendance.
Pulitzer Center and GIJN publish first global AI accountability reporting guide for investigative journalists
The Pulitzer Center's AI Accountability team and GIJN (Global Investigative Journalism Network) jointly published an AI Accountability Reporting Guide on June 22, 2026. Written by Pulitzer Center journalists Gabriel Geiger and Lam Thuy Vo, with contributions from Karen Hao, Laís Martins, and Pablo Jiménez Arandia, the guide gives investigative reporters a four-stage framework for examining AI systems: inputs (data and compute), model training, application deployment, and societal impact. A companion webinar, "Uncovering AI's Human Cost," is scheduled for June 30. GIJN is a global nonprofit serving investigative journalists in more than 115 countries; its network includes member outlets from Turkey, the DRC, and across Latin America.
gijn.org ↗Why it matters
the guide gives journalists everywhere a structured entry point for AI accountability reporting, which means the kinds of questions being asked of AI-deploying organisations are about to become more systematic and better informed. Nonprofit communicators and advocacy teams can use the four-stage framework in reverse to anticipate what questions a well-briefed journalist would now ask about their AI use, and to prepare governance and transparency documentation accordingly.