AI news5
Anthropic releases Claude Fable 5, its most powerful public model yet
On 9 June Anthropic released Claude Fable 5, the first broad release from its top-tier "Mythos" class, calling it the most capable model it has ever made widely available, with a growing lead on long, complex software, knowledge and vision tasks. Safeguards block responses in high-risk cybersecurity, biology and chemistry areas and quietly fall back to Claude Opus 4.8, which happens in fewer than 5 percent of sessions. A restricted twin, Claude Mythos 5, ships to vetted cyber-defence partners with some safeguards lifted, and a biomedical trusted-access program follows in weeks. Pricing is $10 per million input and $50 per million output tokens, with Mythos-class traffic kept under 30-day retention.
theverge.com ↗Why it matters
the release sets a template where capability is no longer the gate, the safeguards are, and that tiered model is what makes frontier AI usable in sensitive settings.
Beacon raises $225M for an "anti-PE" AI roll-up of small software firms
Toronto and San Francisco-based Beacon raised a $225 million Series C led by General Catalyst and HarbourVest, pushing total funding past half a billion dollars in barely two years. Rather than strip costs like a typical private-equity buyer, Beacon buys small software companies, now roughly one a week, and rebuilds them on a shared AI-native platform, automating accounting, payroll and the products themselves. It says the approach has driven more than 50 percent EBITDA growth across its portfolio in the past year. It is the highest-profile name in a fast-spreading "AI-enabled roll-up" thesis.
thenextweb.com ↗Why it matters
it is an early test of whether AI changes who can profitably run small software and services businesses, a model that could spread into professional services.
PhysicsX raises $300M at a $2.4B valuation for AI engineering
London-based PhysicsX raised $300 million at a $2.4 billion valuation, more than double its worth a year ago, led by Temasek with Nvidia, Siemens, Applied Materials and others. Founded by former Formula 1 engineers, it uses AI to compress design and simulation work that once took months into seconds, aimed at manufacturing and defence. The round brings total funding to around $500 million and will fund AI research and US plus Singapore expansion.
tech.eu ↗Why it matters
AI is moving from text and images into physical engineering, where faster simulation reshapes how real-world products and infrastructure get designed.
Google DeepMind, Wellcome Sanger and Google.org form a five-year AI genomics consortium
The Wellcome Sanger Institute, Google DeepMind and Google.org announced a five-year consortium to generate large-scale genomic datasets for the machine-learning models used in biological research, focusing on areas where data is scarce or poorly structured. It pairs Sanger's genomics and data-generation work with DeepMind's AI, while Google.org supports open-access data. No funding figure was disclosed.
edtechinnovationhub.com ↗Why it matters
building open, AI-ready scientific datasets is the unglamorous groundwork that decides whether AI delivers real biomedical breakthroughs or just hype.
Talkspace launches Tee, a safety-gated clinical AI agent for mental health
Behavioural-health provider Talkspace released Tee, a fine-tuned large language model built by mental-health experts as a safer alternative to general chatbots people already use for personal support. It has HIPAA-grade privacy, detects suicide, violence, abuse and seven other risk categories, and provides real-time oversight by licensed clinicians with human intervention when needed. It is available to adults at $19.99 a month after a trial.
businesswire.com ↗Why it matters
as people increasingly take personal crises to general chatbots, purpose-built clinical guardrails and human escalation are becoming the line between helpful and harmful.
AI in the nonprofit sector2
A large anti-poverty nonprofit pledges no AI layoffs and expands reach instead
Met Council, the largest US Jewish anti-poverty nonprofit, says it is the first organisation in the country to tell staff it will not lay anyone off because of AI, framing the technology as a tool rather than a replacement. It built an AI digital food pantry that predicts demand and routes deliveries, plus an in-house intake bot called Reggie that connects clients to the right service, from food assistance to domestic-violence support. CEO David Greenfield says this let Met Council grow its reach from 325,000 to more than 350,000 people without adding to staff workload, and the org pays staff two hours a week to learn AI. Greenfield now sits on New York's AI FutureWorks Commission.
jns.org ↗Why it matters
it is a concrete model of human-centred AI adoption in a nonprofit, expanding service capacity while explicitly protecting the workforce.
A nonprofit's smartphone AI screens for blindness in underserved regions
Visilant, a nonprofit eye-health-tech organisation, is deploying smartphone-based AI to detect eye disease and cut avoidable blindness in remote areas, and is now seeking partners for a Kenya rollout. A clip-on lens and light module turns a budget Android phone into a screening device, with AI flagging likely causes and whether a patient needs referral, and it works offline until a connection is available. A 1,000-patient study at an Indian eye hospital found 96 percent diagnostic agreement with doctors, and health workers screened each eye in under three minutes. Visilant has screened about 50,000 patients in India and uses richer-market sales to subsidise low-income regions.
healthbusiness.co.ke ↗Why it matters
it shows AI extending scarce specialist diagnosis into places with too few clinicians, with clinical validation and human referral built in.