AI news5
Five Eyes intelligence alliance warns AI-enabled cyberattacks are months, not years, away
The US, UK, Canada, Australia, and New Zealand spy agencies issued a joint statement on June 23 warning that frontier AI models capable of launching major cyberattacks against governments and businesses could emerge within months. The agencies urged immediate investment in cyber defences and called out a "massive gap" in preparedness, particularly for small and medium businesses. The statement follows the Trump administration's suspension of Anthropic's Mythos 5 and Fable 5 models for foreign nationals and references the Fable jailbreak concern that triggered that action. The alliance also noted that AI tools are part of the defence: organisations integrating AI into security operations can detect vulnerabilities faster than adversaries.
cp24.com ↗Why it matters
the Five Eyes agencies naming a concrete months-not-years timeline shifts the AI security conversation from long-horizon policy debate to immediate operational risk, directly pressuring every organisation that has underinvested in cyber defences to act now.
Google DeepMind invests $75M in indie studio A24 to build AI filmmaker tools
Google DeepMind announced a $75 million research investment in A24, the production studio behind films including Everything Everywhere All at Once and Marty Supreme, to develop AI-powered filmmaking workflows and techniques. The deal is structured as a research partnership, with DeepMind contributing model capability and A24 providing the creative and production context. It is one of the largest direct investments by a frontier AI lab in a creative sector company. The announcement came June 22.
techcrunch.com ↗Why it matters
a leading frontier research lab partnering with the cultural-credibility studio in independent film signals that AI in creative industries is moving from generic productivity claims to specific, filmmaker-endorsed tools, a reputational anchor that will be used to justify broader creative AI deployment.
Japan's Sakana AI launches Fugu, an orchestration model designed to route around export control disruptions
Sakana AI released Fugu on June 23, an orchestration model that farms each user request out to a pool of underlying models through a single API, hiding the multi-agent architecture from the caller. It comes in two variants: Fugu for everyday coding and chat, and Fugu Ultra for patent research and security testing, with Sakana claiming both perform at or above Fable 5 on several coding and reasoning benchmarks. The lab explicitly pitched Fugu as "delivering frontier capability without the risk of export controls" following the US suspension of Anthropic's top models. Early reception has been mixed, with users and researchers reporting that real-world performance falls short of the benchmark numbers and raising questions about the underlying model mix.
sakana.ai ↗Why it matters
Fugu represents Japan's first direct answer to the export control squeeze on top-tier Western models, taking an orchestration approach rather than building a competing foundation model, which is a structurally different and potentially lower-cost path to frontier-adjacent capability.
SpaceX leases $6.3B of Nvidia compute at Colossus data centres to AI startup Reflection AI
SpaceX signed a deal with Reflection AI on June 22 to rent $6.3 billion of Nvidia computing power from its Colossus data centres in Memphis. Colossus was originally built to train xAI's Grok models but has since pivoted to renting compute to other AI labs, with Anthropic leasing $1.25 billion per month, Google at $920 million, and Cursor acquired entirely for $60 billion. Reflection AI, founded in October 2025 to build open frontier systems for government and enterprise, becomes the fourth major Colossus tenant but has yet to release a public model.
cnbc.com ↗Why it matters
SpaceX converting its own-use AI infrastructure into a compute rental business reveals a second revenue stream that is structurally separate from model performance, making Colossus financially viable even if Grok falls short at the frontier.
ByteDance unveils Seedance 2.5: native 30-second AI video clips at up to 4K resolution
ByteDance announced Seedance 2.5 at the Volcano Engine FORCE conference on June 23, a video generation model capable of producing native 30-second clips at up to 4K resolution. The model supports up to 50 reference inputs, includes local re-draw editing, and targets generation speeds approaching real time. A public launch is expected in early July 2026. The announcement came weeks after ByteDance's Seedance 2.0 went viral in China for Hollywood-level production quality on short clips.
the-decoder.com ↗Why it matters
extending native output from the current 5-10 second standard to 30 seconds at 4K without stitching removes the main workflow friction in AI video production and brings model-native long-form content a step closer to the current human-production floor.
AI in the nonprofit sector4
Cadence raises $100M to expand AI chronic care monitoring into nonprofit health systems
Cadence, which builds an AI clinical intelligence platform for older adults with chronic conditions, closed a $100 million Series C on June 23 led by Spark Capital. The company's Proactive Care Engine analyses patient vitals in real time, alerts clinical teams to emerging risks between visits, and supports medication adjustments through supervised AI agents. New partnerships were announced with Duke Health and Texas Health Resources, a North Texas-based nonprofit health system. Prior health system partners include the American Heart Association, Hartford HealthCare, and Rush University System for Health.
mobihealthnews.com ↗Why it matters
AI-powered remote monitoring reaching nonprofit health systems that serve Medicare and Medicaid patients closes a gap between what large commercial health systems can deploy and what community-anchored nonprofits have had access to, with direct downstream impact on chronic disease outcomes in under-resourced populations.
Petco Love Lost nonprofit AI has reunited 250,000 pets, averaging a new reunion every four minutes
National nonprofit Petco Love released an update on June 23 on its Petco Love Lost platform, which uses AI photo-matching across more than 500 visual markers to identify lost pets from a single uploaded photo. The system searches nationwide shelter databases and community platforms including Nextdoor and Neighbors by Ring automatically. The milestone of 250,000 reunions was reached ahead of National Lost Pet Prevention Month in July, with reunions in some cases happening within 20 minutes of a lost profile being created. The platform is free to use.
prnewswire.com ↗Why it matters
a nonprofit running free AI infrastructure at one reunion every four minutes demonstrates that animal welfare organisations can operate AI systems with measurable, sustained population-level outcomes rather than pilot-stage results.
Catholic health systems Trinity Health, Mercy, and SSM Health are deploying AI for radiology and imaging at scale
The Catholic Health Association's Health Progress Summer 2026 issue documents major Catholic nonprofit health systems moving past AI pilots and into operational deployment for radiology and imaging services. Trinity Health, Mercy, and SSM Health are among the systems cited as having successfully deployed AI-supported tools to augment their radiology workflows. Catholic health systems collectively represent one of the largest nonprofit healthcare delivery networks in the US, with more than 650 hospitals and long-term care facilities across 21 states.
chausa.org ↗Why it matters
when the Catholic health system network, which serves a disproportionately high share of rural and low-income patients, reaches operational AI deployment in diagnostic imaging, the technology's reach into underserved communities scales by millions of patients simultaneously.
AACR presents Cancer Disparities Progress Report 2026 to Congress as nonprofit cancer sector confronts AI equity questions
The American Association for Cancer Research, a nonprofit scientific organisation representing more than 63,000 researchers in 145 countries, released its Cancer Disparities Progress Report 2026 on June 24 at a congressional briefing in Washington DC. The annual report documents persistent gaps in cancer incidence and mortality across race, income, and geography, and this year arrives as AI-powered diagnostic and treatment tools are entering clinical oncology settings at scale. The congressional briefing format, which AACR uses when seeking legislative or regulatory action, signals the organisation is ready to push for policy responses rather than simply documenting the problem.
aacr.org ↗Why it matters
as AI diagnostic models reach clinical oncology, the question of whether they replicate or reduce the outcome gaps documented in this report becomes a policy decision, and a congressional briefing is the formal mechanism through which a research nonprofit puts that question on the legislative agenda.