AI news5
Anthropic signs SpaceX deal for 300MW Memphis Colossus 1 capacity
Anthropic signed an agreement with Elon Musk's SpaceX to access more than 300 megawatts of compute at the Colossus 1 data centre in Memphis. The deal "substantially" expands Claude capacity and lets Anthropic raise usage limits, announced 6 May 2026. It lands the same week Anthropic separately reportedly signed a roughly $200B five-year cloud agreement with Google Cloud.
bloomberg.com ↗Why it matters
Anthropic is locking in cross-vendor compute on a scale that signals frontier-model demand has decoupled from any single cloud, with downstream pricing and availability effects every AI buyer will feel.
OpenAI rolls out GPT-5.5 Instant as the new ChatGPT default
OpenAI began rolling out GPT-5.5 Instant on 6 May 2026, replacing GPT-5.3 Instant as the default model in ChatGPT. The update targets fewer hallucinations, tighter answers, and reduced formatting clutter, with a particular emphasis on finance, law, and healthcare reliability. No new model name, just the swap underneath.
thurrott.com ↗Why it matters
Quiet default-model swaps move the floor on what users experience without any rollout announcement, which is now the dominant pattern for capability gains across the major chat assistants.
Anthropic previews "dreaming" — Claude agents reviewing their own work between sessions
Anthropic unveiled a research-preview feature called "dreaming" that lets Claude agents review their previous sessions to identify patterns and update user-context files for self-improvement. The idea is to compress the gap between human-feedback retraining and runtime adaptation. Released as a research preview on 6 May 2026, no general availability yet.
businessinsider.com ↗Why it matters
Memory-and-reflection between sessions is the missing capability that turns ad-hoc agent tools into compounding workflows, and an Anthropic-shipped version sets the bar for the rest of the field.
Microsoft-OpenAI contractual definition of AGI exposed in Musk trial
Court exhibits in Musk v. Altman revealed the 36-page Microsoft-OpenAI contract definition of AGI: a "highly autonomous system that outperforms humans at most economically valuable work". Reported 6 May 2026 by The Verge. The clause was a contractual ceiling that, if triggered, would alter Microsoft's IP rights and revenue share.
theverge.com ↗Why it matters
A leaked AGI clause forces public scrutiny of how the two largest AI vendors are privately defining the threshold that ends their commercial alliance, which is a governance question regulators and counterparties will now have to engage with directly.
Google DeepMind takes minority stake in CCP Games to use Eve Online for AI training
Google DeepMind disclosed a minority stake in CCP Games, the maker of long-running MMO Eve Online, to use the game environment for AI research and training, reported 6 May 2026. Eve Online's persistent player-driven economy gives DeepMind a live multi-agent environment that pure simulation cannot reproduce.
bloomberg.com ↗Why it matters
Frontier labs are buying access to messy, social, real-stakes environments rather than synthesising them, which shifts where novel agentic capabilities will get tested first.
AI in the nonprofit sector4
OpenAI launches European Youth Safety Blueprint + €500K EMEA grants for AI chatbots
OpenAI announced the European Youth Safety Blueprint on 6 May 2026, a five-pillar framework for protecting young users on AI chatbots covering age assurance, parental controls, and bans on manipulative design. The launch is paired with €500,000 in EMEA grants for nonprofits and researchers working on youth safety, AI literacy, and safe-by-default product design.
edtechinnovationhub.com ↗Why it matters
A frontier-lab policy framework with an attached funder pot in the same announcement is the new playbook, and youth-safety-aligned EMEA nonprofits should treat this as an active funding window.
University of St Andrews wins €1.3M AFRITRAC grant to use AI for African infectious-disease prediction
The University of St Andrews School of Medicine secured a €1.3 million international grant for AFRITRAC, a four-year project running May 2026 to April 2030. The project strengthens Africa's capacity to predict and respond to infectious disease outbreaks using AI, big data, and locally trained data scientists. Partners span six African research institutions.
news.st-andrews.ac.uk ↗Why it matters
AI-for-public-health is moving from frontier-lab demos into multi-year capacity-building grants run with African institutions as principal investigators, which is the model donors will increasingly fund over single-vendor procurements.
Virginia State University awarded $1.03M federal grant for Center for Generative AI and Industrial Cybersecurity
Virginia State University, an HBCU, received $1.03 million in federal funding on 5 May 2026 to establish the Center for Generative AI and Industrial Cybersecurity. The centre will study misinformation, bias, copyright, and labour-market AI risks while preparing students for an AI-affected workforce. Federal funding source is tied to industrial-cybersecurity workforce development.
vsu.edu ↗Why it matters
HBCU-led AI research centres receiving federal seed funding shifts the geography of who gets to shape AI risk discourse, an angle nonprofits working on equity-led AI policy can build coalitions around.
Fundory launches AI-powered grant-writing platform aimed at nonprofits and small charities
Fundory, built on Brownmine AI, launched on 1 May 2026 as a grant-writing platform targeting nonprofits, small businesses, and community organisations. The platform identifies relevant funding opportunities, drafts tailored proposals, and optimises submissions, positioned against the legacy spreadsheet-and-Word workflow most small NPs still run.
einpresswire.com ↗Why it matters
Grant-writing is the highest-leverage AI use case for under-resourced nonprofits, and a vendor explicitly targeting that workflow is worth tracking against the in-house Claude/ChatGPT pattern most operators currently use.