AI news5
Anthropic launches Claude Security public beta with Opus 4.7, CrowdStrike + Microsoft Security + Palo Alto Networks + SentinelOne + Wiz embedded as partners
Anthropic opened Claude Security to public beta on 30 April for Claude Enterprise customers, with Team and Max access following soon. The product runs on Opus 4.7 to scan code for vulnerabilities and propose patches, accessible from the Claude.ai sidebar or claude.ai/security with scheduled scans, audit-system integration, and triaged-finding tracking. Technology partners CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, TrendAI and Wiz are embedding Opus 4.7 in their tools, while Accenture, BCG, Deloitte, Infosys and PwC ship Claude-integrated security services. The release follows hundreds of organizations testing it during a limited research preview.
claude.com ↗Why it matters
Vulnerability scanning is now distributed across the major enterprise security stack overnight, and any nonprofit running on Microsoft Defender or CrowdStrike Falcon will inherit Opus 4.7-powered scanning whether or not they directly procure Claude.
Anthropic moves to close ~$50B financing round at $850-900B valuation while run rate approaches $40B and $1M+ enterprise customers cross 1,000
TechCrunch reported on 29 April that Anthropic is closing a roughly $50 billion preemptive round at an $850-900 billion valuation, only weeks after the $1 trillion milestone reframed last week. Run rate is now near $40 billion, up from approximately $9 billion at the end of 2025, with the company adding more than 1,000 enterprise customers each spending at least $1 million annually, a count that doubled in under two months. Saastr separately reported Anthropic crossing $14 billion in true ARR, up from $1 billion 14 months earlier.
techcrunch.com ↗Why it matters
The growth is now driven by enterprise commitments large enough that procurement teams in nonprofits with even modest Claude usage should expect their account terms, support tier, and invoicing posture to shift inside the next 12 months.
Perplexity Computer launches enterprise workflow templates plus prebuilt connectors to Snowflake, Datadog, Salesforce, SharePoint, HubSpot and any custom MCP server
Perplexity expanded its Computer agent product on 30 April with workflow templates for legal contract review, finance audit support, sales call preparation and customer support triage, alongside business-grade connectors to Snowflake, Datadog, Salesforce, SharePoint and HubSpot. Pro, Max and Enterprise users can plug any external tool into Perplexity through the Model Context Protocol with OAuth, API key or open authentication, on top of more than 400 prebuilt curated connectors. Enterprise tier ships SOC 2 Type II, SSO/SAML, SCIM provisioning, granular admin controls and full audit logging.
perplexity.ai ↗Why it matters
A nonprofit running Salesforce NPSP and SharePoint can now point Perplexity at its operational stack without dev work, which is the first time a frontier search-and-agent vendor has matched ChatGPT for Business and M365 Copilot on out-of-box NGO-tooling reach.
OpenAI publishes "Where the goblins came from" tracing GPT-5.1's metaphor quirk to personality-tuning reward signals
OpenAI released a public technical post on 1 May explaining how heightened use of "goblin"-style metaphors in GPT-5.1 traced back to specific reward signals during personality tuning. The piece walks through how small incentive choices in post-training cascade into observable model behaviour at deployment scale. The disclosure follows a leaked Codex system prompt the previous week instructing the model to "never talk about goblins," which puzzled researchers until this explanation landed.
openai.com ↗Why it matters
This is a concrete worked example of why nonprofit teams cannot treat tone or persona issues as cosmetic, because the same signal that makes a model overuse one metaphor also explains why a fundraising assistant might suddenly hedge or moralise against a donor's wishes.
Samsung Q1 net profit clears $30 billion as AI memory super-boom rewires the semiconductor industry
Samsung reported a first-quarter net profit equivalent to more than $30 billion on 30 April, blasting past its previous quarterly record and almost matching its prior full-year high. The Wall Street Journal reported memory-chip pricing has entered a structural super-boom driven by AI training and inference demand, with the supply crunch expected to deepen through 2027. Memory now sits among the most profitable products in the world, displacing prior peaks set in oil and pharmaceuticals.
wsj.com ↗Why it matters
AI inference cost curves stop falling as fast when memory costs climb, and any nonprofit budget built on the assumption that "AI gets cheaper every year" needs a 2026-2027 memory-pricing line item to stay credible.
AI in the nonprofit sector4
Wikimedia Enterprise expands paid AI-content partnerships with Amazon, Meta, Perplexity, Mistral and others on Wikipedia's 25th birthday
The Wikimedia Foundation announced new Wikimedia Enterprise partnerships with Amazon, Meta, Perplexity, Mistral AI and several other AI companies on 27 April, timed to Wikipedia's 25th anniversary. The partners get content access at the volume and speed they need for AI training and inference, while paying fees that flow back into the nonprofit's sustainability funding. The Foundation also published its draft FY 2026-2027 annual plan on the same date, framing this revenue as core to sustaining open knowledge in an AI era.
diff.wikimedia.org ↗Why it matters
This is the cleanest live model for any nonprofit sitting on a corpus of mission-relevant content asking how to charge AI labs without compromising open access, and the Wikimedia Enterprise terms become a useful procurement reference point regardless of whether the org licenses Claude or ChatGPT internally.
NC State publishes open-access food-bank routing optimisation framework built specifically for nonprofit hunger-relief operations
NC State engineering researchers published an open-access paper in Computer-Aided Civil and Infrastructure Engineering in April describing an Anticipatory Monte Carlo Tree Search framework for stochastic dynamic routing of food deliveries. The model handles uncertain volunteer availability, dynamic household requests and limited transport capacity, all of which break conventional commercial routing software. An app built on the framework is in development with the goal of free distribution to nonprofit food banks.
news.ncsu.edu ↗Why it matters
Food banks running on donated trucks and volunteer drivers have been bracketed out of every commercial route-optimisation product, and a free, peer-reviewed nonprofit-specific tool removes the most common excuse for not modernising the most logistics-heavy nonprofit operation.
NTEN AI for Nonprofits Tech Readiness Cohort opens for free US nonprofit places with bundled Anthropic tool access
NTEN's AI for Nonprofits: Nonprofit Tech Readiness cohort is taking applications for a free 20-seat program for US nonprofits, designed for staff in any role wanting to build AI deployment capacity for their org. Participants get hands-on coaching plus free access to Anthropic tools through a partnership announced for the cohort. The cohort sits alongside 27NTC early-bird registration which is open through 31 May 2026.
nten.org ↗Why it matters
Any AIH member or AI-curious nonprofit operator who has been hesitating on Claude pricing now has a US-only no-cost path to a curated cohort with the model bundled in, and the application window will close fast given the 20-seat cap.
Zendesk Tech for Good Impact Awards 2026 open for global nonprofits with $5K-$50K grants and AI Resolution Platform access
Zendesk's Tech for Good Impact Awards 2026 is taking applications globally with a 15 July 2026 deadline, awarding $5,000 to $50,000 in cash plus in-kind access to Zendesk's AI-powered Resolution Platform. Eligibility is restricted to registered nonprofits not currently using Zendesk and that have not received prior Zendesk grants, with focus areas covering community connection, crisis resilience and tech-career pathways. Geographic exclusions list 18 countries including Russia, Iran and North Korea.
techforgood.zendesk.com ↗Why it matters
For any nonprofit running a service desk or member-support function on a basic shared-inbox setup, this is one of the few AI-tooling grants that pairs cash with the actual platform access, removing the usual gap between "we got a grant" and "we deployed something."