AI news5
OpenAI previews GPT-5.6 Sol, staggers public launch at Trump administration's request
OpenAI published a preview announcement for GPT-5.6 Sol on June 26, describing a next-generation model with significant capability advances, but confirmed initial access is limited to a small group of government-approved partners. The restriction follows a direct request from the Trump administration, which asked OpenAI to hold back broad rollout while the government vets frontier AI deployments under the June 2 executive order's 30-day review framework. OpenAI published a GPT-5.6 Preview System Card simultaneously on deploymentsafety.openai.com. Full public access is expected within weeks.
openai.com ↗Why it matters
a US president dictating the pace of a commercial AI release is a new precedent, and the system card publication alongside a limited rollout sets a template for how frontier model governance may work under the current administration, with direct implications for any nonprofit or public sector org planning to build on GPT-5.6 APIs.
AI framework discovers GPNMB CAR-T cells as a multi-cancer therapy in Cell journal
Penn researchers published a study in Cell on June 25 showing that an AI framework can identify promising CAR-T cell therapy targets in a matter of weeks rather than months. The AI screened candidate antigens and ranked GPNMB as its top hit, and subsequent preclinical testing confirmed that GPNMB-directed CAR-T cells showed robust efficacy across melanoma, leukemia, and colorectal cancer models. The paper suggests the same AI-guided discovery process is transferable to other cancer types and represents a new pipeline for identifying shared tumour antigens.
cell.com ↗Why it matters
compressing target discovery from months to weeks is the kind of compounding efficiency gain that reshapes clinical trial timelines, and a multi-cancer CAR-T candidate found via AI signals that the bottleneck in oncology is shifting from computing power to biological validation.
Samsung Group unveils $646 billion, ten-year AI and semiconductor spending plan
Samsung Group is set to announce a 1,000 trillion won ($646 billion) investment package covering semiconductors, AI data centers, batteries, and displays over the next decade, reported by Bloomberg on June 26. More than 350 trillion won ($226 billion) is earmarked specifically for AI data center facilities, meaning Samsung intends to own and operate the infrastructure its chips run in, not just manufacture components. SK Hynix is reportedly preparing a comparable announcement, indicating coordinated Korean industrial policy around AI infrastructure dominance.
bloomberg.com ↗Why it matters
the scale signals that AI compute infrastructure is entering a phase of state-adjacent strategic investment, where the companies building it are making 10-year capacity bets that will shape the cost and availability of AI services globally for the rest of the decade.
Unconventional AI releases Un-0, the first model built on oscillator-based computing architecture
Naveen Rao, formerly Databricks' head of AI, launched Unconventional AI and on June 25 published Un-0, an image-generation model built on a software simulation of a new oscillator-based chip architecture rather than conventional GPUs. The company claims the architecture can reduce AI inference power consumption by up to 1,000 times relative to current state-of-the-art hardware, with Un-0 showing comparable image quality to leading diffusion models in the simulation. Unconventional AI plans to release actual chip schematics soon and build a full inference stack, eventually offering compute capacity as a provider.
techcrunch.com ↗Why it matters
if the power-efficiency claims hold at hardware scale, oscillator-based computing would fundamentally change the energy economics of AI deployment and is the most credible hardware-level challenge to GPU-dominant inference since the NPU generation began.
California becomes the first state to launch a tool tracking AI-related job loss
Governor Newsom announced the California AI-Unemployment Tracker (CAIT) on June 25, a public dashboard that links monthly unemployment insurance claims to measures of occupational AI exposure, enabling researchers and policymakers to monitor which job categories are experiencing AI-correlated layoffs. The tool is the first state-level instrument in the US to connect workforce data to AI risk metrics rather than relying on self-reported displacement figures. Initial data released with the tracker shows no statewide surge in AI-attributed unemployment yet, but establishes the measurement infrastructure to detect trends as they emerge.
gov.ca.gov ↗Why it matters
workforce organisations, nonprofits, and community foundations working on economic mobility now have a publicly-accessible, government-maintained data signal to validate where AI displacement is actually happening versus where it is projected, which changes how programs are targeted and how funding cases are made.
AI in the nonprofit sector4
RAISE US launches with $500M+ to retrain American workers displaced by AI
Former Commerce Secretary Gina Raimondo and former Indiana Governor Eric Holcomb launched RAISE US on June 25, a nonpartisan national nonprofit designed to help the US workforce navigate AI-driven job transitions. The organization launches with more than $500 million committed toward a $1 billion target, backed by anchor partners OpenAI, Anthropic, Amazon, and Microsoft alongside Bank of America, General Motors, Blackstone, Deloitte, IBM, and more than 20 other companies and foundations. Initial state partnerships in Arkansas, Connecticut, Maryland, and Utah will pilot outcome-linked workforce training, corporate reskilling incentives, and earn-and-learn apprenticeships. Arkansas LAUNCH, an AI-powered career navigation platform, is the first active deployment under the RAISE US model.
rockefellerfoundation.org ↗Why it matters
RAISE US is the first initiative to seat the AI companies building the technology at the same table as the companies deploying it and the states responsible for workforce programs, and the $500M+ launch with bipartisan governors backing creates a template for AI workforce transition that will be studied and replicated.
Kenya's government opens exploratory talks with Anthropic on AI for health, education, and food security
Ambassador Philip Thigo, Kenya's Special Envoy on Technology, held a series of discussions with Anthropic's Elizabeth Kelly, head of the company's Beneficial Deployments team, on June 26. The conversations covered four development priorities: preserving and expanding access to Kenya's more than 40 languages, strengthening pandemic preparedness and disease surveillance, advancing food systems innovation, and improving learning outcomes through AI-powered education tools. No partnership or formal agreement was announced; both sides described the engagement as productive and expressed interest in continuing discussions.
cioafrica.co ↗Why it matters
Kenya is simultaneously in talks with OpenAI (President Ruto met Sam Altman at the G7 in June) and now Anthropic, positioning it as the most active African government negotiating direct AI partnerships with frontier labs for public-sector deployment, with health and education as the named priorities.
Open Source Initiative launches two-year AI fellowship at UN Open Source Week
The Open Source Initiative (OSI) announced the Open Source AI Fellowship on June 23 at UN Open Source Week in New York, backed by Duke University's Sanford School of Public Policy alongside Red Hat, AWS, Google, Automattic, and Mozilla. Gabriel Toscano, a freshly graduated Master of Public Policy from Duke, will serve as inaugural fellow, with a first-year research agenda covering the value created by open-source AI systems, analysis of models released under the Open Source AI Definition (OSAID) since version 1.0, and multi-stakeholder convenings to build shared understanding. The launch follows the G7 adopting OSI's AI openness framework in May and the EU's June Tech Sovereignty package incorporating OSI recommendations on procurement and standards.
prnewswire.com ↗Why it matters
as governments and foundations increasingly make open-source AI a procurement and governance preference, the OSI fellowship becomes the body defining what "open source" actually means for AI systems, which directly affects which tools nonprofits can adopt with confidence under funders' open-source requirements.
Tempus AI and Angiosarcoma Awareness partner on 600-record multimodal cancer dataset
Tempus AI and nonprofit public charity Angiosarcoma Awareness, Inc. announced a research collaboration on June 25 to study the biology of angiosarcoma, a rare and aggressive blood vessel cancer. The partnership will draw on Tempus' dataset of approximately 600 de-identified angiosarcoma records, each containing paired DNA and RNA sequencing data, along with the company's computational and analytical infrastructure. The collaboration is designed to uncover biological insights that could inform future treatment development for a cancer type with very limited research data relative to its severity.
tempus.com ↗Why it matters
this is a direct model for how disease-specific nonprofit patient advocacy organizations can leverage AI company data partnerships to accelerate research on rare conditions that lack the patient volume to attract pharmaceutical investment on their own.