AI news6
PwC commits to train 30,000 US staff on Claude in expanded Anthropic alliance
PwC and Anthropic announced on 14 May an expanded alliance in which PwC will train and certify 30,000 US professionals on Claude Code and Claude Cowork before rolling the deployment to its 364,000-strong global workforce across 136 countries. The deal establishes a joint Center of Excellence and a Claude-native finance group inside PwC's Office of the CFO. PwC reports Claude is already cutting delivery times by up to 70 percent across professional sports operations, insurance underwriting, mainframe modernisation, HR and cyber, with one client compressing insurance underwriting from 10 weeks to 10 days.
anthropic.com ↗Why it matters
Big Four professional services firms function as enterprise distribution channels for AI models. This is the largest single Claude rollout to date and a direct test of whether Anthropic can convert its business adoption lead over OpenAI into durable revenue.
Anthropic Pentagon designation now appearing as risk factor in tech filings
Bloomberg reported on 14 May that Anthropic's "supply chain risk" designation by the Pentagon is now flagged in SEC filings from companies including Figma and Freightos, who say a federal ban on Claude could materially damage their government business. Figma stated its AI features for federal agencies are built on Claude and a forced replacement could hurt sales to government and regulated buyers. Anthropic is suing the Defense Department, arguing the designation stems from its refusal to allow certain military uses of its models.
bloomberg.com ↗Why it matters
A single AI provider's political fight with Washington is now a contagion risk for any company that built features on top of Claude. Procurement teams across the AI supply chain will start writing fallback clauses into contracts.
Federal judge pauses Anthropic $1.5bn authors settlement, asks for fee detail
At a 14 May San Francisco hearing, US District Judge Martinez-Olguin declined final approval of Anthropic's $1.5 billion settlement with authors over training-data copyright, asking lawyers to file written supplements on fee structures and lead-plaintiff payments. The settlement covers more than 480,000 works, with claims already filed by rights-holders of over 92 percent of them, and would be the largest copyright settlement in US history. A group of 25 writers including Dave Eggers opted out and filed a separate California complaint the same week.
reuters.com ↗Why it matters
This is the first major US AI-training copyright case to reach settlement. How the court rules on fees and payouts sets the template every other AI lab will negotiate against, with OpenAI, Meta and Google all facing similar suits.
Microsoft scouting AI startups for acquisition, weighing Inception to ease OpenAI dependence
Reuters reported on 14 May that Microsoft has begun systematically scouting AI startups to acquire, with the goal of building in-house frontier models and reducing its OpenAI reliance. Live targets include Inception, a Stanford-founded lab building diffusion-based language models that run roughly 10 times faster than autoregressive LLMs and is now valued above $1 billion. Microsoft's M12 venture arm invested in Inception's $50 million seed round in late 2025. Earlier Microsoft talks with Cursor reportedly stalled on antitrust concerns over overlap with GitHub Copilot.
reuters.com ↗Why it matters
The Microsoft-OpenAI partnership is the most consequential commercial relationship in AI. If Microsoft starts buying its way to model independence, it reprices Azure capacity, OpenAI revenue floor, and the entire frontier lab market.
IBM launches Forward Deployed Units, six-person agent pods for enterprise AI delivery
IBM Consulting on 14 May unveiled Forward Deployed Units, a delivery pod combining senior business specialists, architects, and engineers with a digital workforce of specialised agents handling coding, evaluation, testing, and documentation. IBM claims a six-person FDU pod delivers the output of a 30-person team. The model is already live at Riyadh Air, Nestlé, Heineken, and Pearson, and is being deployed globally across Asia Pacific, Europe, and the US through IBM Consulting Advantage.
newsroom.ibm.com ↗Why it matters
This formalises what most large consultancies are already trying to do informally: sell smaller teams plus agent labour. If FDU economics hold up, it changes how every Big Four firm prices AI implementation work.
Verizon joins Anthropic Project Glasswing critical-infrastructure security coalition
Verizon on 15 May confirmed it has joined Project Glasswing, gaining access to Claude Mythos Preview to harden its networks against AI-enabled cyber attacks. The coalition now includes AWS, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, backed by Anthropic's $100M usage credit commitment and a separate $4M in donations to open-source security organisations. Mythos has already identified thousands of high-severity vulnerabilities in major operating systems and browsers.
anthropic.com ↗Why it matters
Telecoms underpin emergency response, banking, and humanitarian comms. Verizon joining shifts Glasswing from a Big Tech security club to genuine critical-infrastructure coverage, with direct implications for nonprofit and government cyber posture.
AI in the nonprofit sector4
Anthropic-Gates partnership names polio, HPV, and pre-eclampsia as first AI-accelerated vaccine targets
Implementation detail surfaced 14 to 15 May for the Anthropic and Gates Foundation four-year, $200 million partnership: the first wave of work will use Claude to computationally screen vaccine candidates for polio, HPV, and pre-eclampsia before pre-clinical development. The largest share of the partnership is targeted at low and middle income countries, where 4.6 billion people lack essential health services. Anthropic and Gates will also work with governments to use Claude on health data for faster decision-making.
gatesfoundation.org ↗Why it matters
This moves the Anthropic-Gates announcement from generic AI for global health framing into specific disease targets with measurable timelines. It is the first large-scale test of using a frontier LLM as front-end triage for vaccine development pipelines in neglected diseases.
Data and Society wins $2M from Humanity AI to build AI Civics public-education program in libraries
Detail emerging from the Humanity AI grant round: Data and Society will receive $2 million of a $3 million AI Civics allocation to build a community-based public education program through libraries nationwide, with the Digital Public Library of America as anchor partner. The program treats AI as a domain of civic life rather than a tool to master, focused on giving communities formal channels into AI design and deployment decisions. Student Defense separately received $500,000 to develop AI adoption guidance for under-resourced higher education institutions through its SHAPE AI initiative.
datasociety.net ↗Why it matters
Library-based AI civics is a genuinely novel delivery model for public AI literacy, hitting communities that NTEN and OpenAI Academy do not reach. If it works, it becomes a template for state and federal AI literacy spend.
AMA tells Congress to add federal guardrails on AI mental-health chatbots as 16 percent of US adults now use them
On 14 May the American Medical Association told Congress that AI chatbots used for mental health care lack consistent safeguards against emotional dependency, misinformation, and inadequate crisis response, and proposed four federal-level requirements including mandatory disclosure, crisis-handoff standards, and advertising restrictions. The AMA cited new KFF polling showing 16 percent of US adults have used AI for mental health support in the past year, and Mpathic research showing leading chatbots still miss subtle long-conversation risk signals.
ama-assn.org ↗Why it matters
Frontline mental-health nonprofits using or building AI tools should expect downstream compliance and disclosure requirements. The ask accelerates the wedge between clinically-supervised tools (Spring ACT, Wysa, Jimini Sage) and consumer chatbots being used off-label.
Anthropic kicks off 10-city free small-business AI tour in Chicago, 100 leaders per stop
Starting 14 May in Chicago, Anthropic and partner Tenex.co launched a free half-day AI fluency tour for small business owners, with 100 attendees per stop receiving a one-month Claude Max subscription. Spring stops are Chicago, Tulsa, Dallas, Hamilton Township, Baton Rouge, Birmingham, Salt Lake City, Baltimore, San Jose, and Indianapolis. Additional cities are confirmed for autumn. The program runs alongside Claude for Small Business, which ships 15 prebuilt agentic workflows across finance, ops, sales, marketing, HR, and customer service.
anthropic.com ↗Why it matters
Free in-person AI training at this scale is rare for solo and small operators, the segment nonprofit capacity-building programs usually miss. Worth flagging to AIH members who run small social enterprises in any of these cities.