AI news5
Anthropic releases Claude Opus 4.8, with Mythos-class cyber models flagged for "coming weeks"
Anthropic released Opus 4.8 on 28 May, an upgrade to its flagship model that it says leads on agentic coding, reasoning, financial analysis, and knowledge work at the same price as Opus 4.7. The company says the new model is about four times less likely than 4.7 to leave flaws in its own code unflagged. Anthropic also said it has made "swift progress" on stronger safeguards and plans to widely release Mythos-level cybersecurity models, previously held back as too dangerous, in the coming weeks.
techcrunch.com ↗Why it matters
a price-flat capability bump plus a near-term general release of frontier cyber models signals Anthropic is moving its most sensitive capability from restricted access toward the open market.
Google Cloud launches AI Threat Defense, bundling Wiz, CodeMender, Gemini, and Mandiant
Google Cloud introduced Google AI Threat Defense on 27 May, an autonomous platform that continuously discovers, prioritises, and remediates software vulnerabilities. It combines four assets: Wiz (the $32B acquisition Google just closed), CodeMender (a DeepMind code-security agent), Gemini for reasoning and code generation, and Mandiant for threat intelligence. It is aimed at enterprises facing attackers who now find and exploit flaws in hours, with launch partners including Accenture, Deloitte, and PwC.
helpnetsecurity.com ↗Why it matters
the big clouds are racing to ship autonomous defensive AI as fast as offensive AI is maturing, and the gap between the two is now the whole security story.
Robinhood gives AI agents their own trading wallets and a virtual credit card
Robinhood announced on 27 May that customers can deploy AI agents to trade stocks and make purchases, connected through its Model Context Protocol service. Agents get a separate account with a dedicated, pre-loaded wallet so they can only spend a capped balance, plus a deletable virtual credit card with monthly spend limits and transaction alerts. The agentic trading feature launches in beta for stocks first, with options, crypto, and futures planned next.
techcrunch.com ↗Why it matters
putting hard spend caps and a disposable card around an autonomous agent is an early, concrete template for how consumer platforms intend to contain agent-initiated money movement.
Bret Taylor's Sierra raises $950M at a $15B valuation for enterprise AI agents
Sierra, the enterprise customer-service AI agent company co-founded by Bret Taylor, raised an additional $950 million at a $15 billion post-money valuation, led by GV and Tiger Global with Benchmark, Sequoia, and Greenoaks joining. The raise comes just eight months after a $350 million round. Sierra says its software is now used by nearly half the Fortune 50 and generates around $150 million in annual recurring revenue.
techcrunch.com ↗Why it matters
capital is concentrating fast in the few agent companies with real enterprise revenue, widening the gap between proven agentic platforms and the long tail of demos.
Chan Zuckerberg Biohub ships a protein "world model" mapping 6.8 billion proteins
The nonprofit Chan Zuckerberg Biohub released a protein-biology "world model" on 27 May, the first major deliverable from the $500M Virtual Biology Initiative it announced in April. It includes ESMC (a protein language model trained on ~2.8 billion sequences), ESMFold2 (structure prediction), and ESM Atlas, a navigable map of 6.8 billion protein sequences and 1.1 billion predicted structures. All three are free to the global scientific community and aim to compress years of protein research into hours or days.
axios.com ↗Why it matters
a nonprofit research institute, not a commercial lab, just shipped one of the largest open AI resources for disease research, keeping a critical scientific layer in the public commons.
AI in the nonprofit sector4
Google.org backs nature nonprofits with the open SpeciesNet model and a $3M Brazil fund
Google.org recently launched a set of nature-focused AI initiatives: SpeciesNet, an open-source model that identifies animal species from camera-trap photos, a $3 million commitment routed through iCS to support AI-enabled solutions from Brazilian nonprofits, and a nature-tech accelerator for startups. The funding supports biodiversity work tied to the global "30x30" goal of protecting 30 percent of land and ocean by 2030. SpeciesNet is freely available for conservation groups to run on their own camera-trap data.
blog.google ↗Why it matters
open species-identification models hand small conservation nonprofits a capability that used to require a data-science team, which is exactly where AI changes what a thin field operation can do.
Opportunity International runs a WhatsApp farm-advisory chatbot for Malawian smallholders
Opportunity International has deployed UlangiziAI, a multilingual WhatsApp chatbot that gives smallholder farmers real-time agricultural guidance in English and Chichewa. It draws on agricultural extension material to answer practical crop and livestock questions for farmers who rarely have direct access to an extension officer. The tool meets farmers on the messaging platform they already use rather than asking them to adopt a new app.
omdena.com ↗Why it matters
extension advice in a local language over WhatsApp is a low-friction frontline model that other agriculture and livelihoods nonprofits can copy without new hardware.
Sonu Sood's charity foundation partners with NPSM to digitise Indian nonprofits
The Sonu Sood Charity Foundation has partnered with NPSM, an AI-native platform from Advance Solutions Corp, to strengthen the digital infrastructure of nonprofits across India, announced in early May. The arrangement includes a technology grant that gives the foundation access to NPSM's AI-native nonprofit operations platform. The aim is to lift the digital backbone of grassroots Indian organisations that have limited in-house tech capacity.
business-standard.com ↗Why it matters
AI-native operations platforms reaching grassroots Indian nonprofits is a different adoption path than the enterprise tools dominating the sector, and India is a large proving ground.
Gates Foundation opens an AI for Giving Grand Challenge with grants up to $150K
The Gates Foundation is running an AI to Accelerate Charitable Giving Grand Challenge, offering grants of up to $150,000 for projects that use AI to help donors give more, and give sooner. The call seeks tools that improve donor engagement, strengthen charitable ecosystems, and increase giving efficiency. It is aimed at social-impact organisations and innovators building practical AI giving tools.
ictworks.org ↗Why it matters
dedicated funding for AI-in-fundraising signals funders now treat giving infrastructure, not just programme delivery, as a place AI should be applied.