AI News Digest: Friday, May 29 2026

Summary for today

Anthropic dominates headlines with a $65B Series H at a ~$965B valuation, $47B annualized run-rate revenue, and a new Claude Opus 4.8 release — signaling the company is approaching both profitability and IPO readiness.
The internet's infrastructure layer is being actively rebuilt for machine traffic: AWS solved a key data center networking problem, Cloudflare and others are redesigning cloud for AI agents, and Google Pay is overhauling its payment rails for autonomous agent transactions.
AI agentic software is hitting genuine product-market fit — Cognition raised $1B at a $26B valuation for its Devin coding agent, while Glean crossed $300M ARR by positioning itself as an AI cost-cutter inside enterprises.
LLM reliability concerns surface on two fronts: research shows models persist in believing false statements even after explicit warnings, and a supply-chain-style prompt injection was hidden in open-source code to sabotage AI coding agents.
On-device and efficient AI models are advancing fast: Liquid AI's MoE model runs 128K context with only 1.5B active parameters, and Perplexity open-sourced a tokenizer with 5× lower latency — both pushing capable AI onto consumer hardware.
Public sentiment toward AI shows cracks: MIT Technology Review's Hype Index highlights graduates booing AI speakers, while returning mothers find software workplaces unrecognizably transformed by AI tooling.

Funding & Valuations

Anthropic raises $65 billion, nears $1T valuation ahead of IPO — At $965B post-money and $47B annualized revenue, this is likely Anthropic's final private round before an IPO that would make it one of the largest tech listings in history.
Anthropic's run-rate revenue hits $47 billion — The revenue figure embedded in the fundraise announcement is the real signal: Anthropic may be on the verge of its first profitable quarter, validating genuine enterprise product-market fit.
I think Anthropic and OpenAI have found product-market fit — Companies paying $200+/month per user for coding and general-purpose agents are covering AI lab costs far better than consumer subscriptions ever could.
[[AINews] Cognition raises $1B in $26B Series D](https://www.latent.space/p/ainews-cognition-raises-1b-in-26b) — Coding agents are being valued as an uncapped TAM play; Cognition's Devin has already cut project timelines for clients like Mercedes-Benz and Itaú.
More Devins in More Places — Cognition will use its $1B raise to expand Devin's reach and improve task-to-model routing, doubling down on autonomous software engineering at scale.
Glean's top line crosses $300M as AI budget-cutting becomes its major selling point — Tripling revenue while selling AI as a cost-reduction tool — rather than a feature — is a durable enterprise positioning that has let Glean survive hyperscaler competition.

Model Releases & Research

Anthropic Ships Claude Opus 4.8 Alongside Dynamic Workflows and Cheaper Fast Mode — Dynamic multi-agent workflows capped at 1,000 subagents and a cheaper fast mode make Claude Code meaningfully more practical for production agentic pipelines.
Claude Opus 4.8: "a modest but tangible improvement" — Anthropic's unusually candid self-assessment of Opus 4.8 as an incremental step stands out in an industry prone to overclaiming benchmark superiority.
Liquid AI Releases LFM2.5-8B-A1B: An On-Device MoE Model — Activating only 1.5B of 8.3B parameters while supporting 128K context, tool calling, and reasoning on consumer hardware pushes capable MoE models decisively onto edge devices.
Perplexity AI Open-Sources Unigram Tokenizer That Achieves 5x Lower p50 Latency — A 5–6× reduction in CPU tokenization overhead is a meaningful production win for high-throughput inference systems, and open-sourcing it benefits the entire ecosystem.
LLMs believe false statements even after explicit warnings that they're false — Fine-tuning experiments revealing a systematic bias toward confidently representing false claims as true pose serious reliability concerns for enterprise deployments relying on LLM reasoning.
Biohub releases a world model of protein biology — The open release of ESMC and ESMFold2 gives the global research community state-of-the-art tools for protein structure prediction and design, potentially accelerating drug discovery significantly.
ElevenLabs Music Generation Model — Music v2's ability to shift genres mid-track while preserving vocal coherence marks a qualitative leap in generative audio, raising the bar for AI-native music production tools.

Infrastructure & Agentic Web

The internet is being rebuilt for machines — The shift from human-centric to machine-dominant internet traffic is prompting fundamental redesigns of cloud infrastructure, APIs, and protocols across AWS, Cloudflare, and others.
Amazon Thinks the Future of Data Centers Depends on a Technical Problem It Just Solved — Amazon's networking breakthrough — dramatically accelerating intra-datacenter information flow — could unlock new performance ceilings for large-scale AI training and inference workloads.
Google Pay preps for AI agents with Universal Commerce Protocol — Redesigning payment infrastructure around autonomous agent transactions rather than human clicks is a foundational step toward a commercially functional agentic internet.
The Age of Async Agents — Cognition's Walden Yan & OpenInspect's Cole Murray — Insights on 80% autonomous Devin commits, spec-to-PR workflows, and full VM environments reveal how production coding agents are already operating well beyond chat-style interactions.
Just like gold and oil, we'll soon be able to trade AI token futures — Treating AI compute tokens as a commodity asset class tradable on derivatives exchanges would fundamentally change how enterprises hedge AI cost exposure and how compute capacity gets allocated.
Reachy Mini goes fully local — Running a robot's conversational AI entirely on-device without cloud dependency is a significant milestone for privacy-preserving, latency-sensitive physical AI applications.

Security & AI Risks

Fed up with vibe coders, dev sneaks data-nuking prompt injection into their code — A hidden instruction in the jqwik open-source library designed to make AI coding agents delete output files reveals an emerging supply-chain attack surface that AI-assisted development uniquely enables.
Cars are trying to spy on you, and it's only just the beginning — As vehicles become AI-data collection platforms, the gap between what automakers harvest and what drivers understand or consent to is widening rapidly.
The AI Hype Index: AI gets booed in graduation season — Class-of-2026 graduates audibly rejecting AI cheerleading from senior tech executives signals that public trust and elite worker enthusiasm for AI are not keeping pace with investment narratives.

Tools, Products & Workplace

Microsoft 365 Copilot gets a speed boost and cleaner design — Doubling load speed and introducing more scannable structured responses addresses the two most-cited friction points in enterprise Copilot adoption.
Asana acquires no-code agent-builder StackAI — Embedding a no-code agent builder directly into a project management platform lowers the barrier for non-technical teams to deploy AI workflows without engineering support.
New Moms Are Returning to Coding Jobs Radically Reshaped by AI — Even a few months of parental leave is now enough to return to a materially different software development environment, highlighting how rapidly AI is restructuring knowledge work norms.
Here Comes Ojai, Waymo's New Chinese-Made Robotaxi — Deploying a Chinese-manufactured vehicle platform in California and Arizona marks Waymo's most direct exposure yet to geopolitical supply-chain scrutiny, even as it expands commercial coverage.
NBA plans AI system for automatic out-of-bounds calls — Adopting a Hawk-Eye-style AI officiating system for possession decisions would be the NBA's most consequential step yet in replacing human judgment with automated calls in live competition.

Developer Resources & Data Science

The Infrastructure Behind Making Local LLM Agents Actually Useful — Practical lessons on combining vLLM with long-context infrastructure for scientific agents offer a replicable blueprint for teams moving beyond hosted API dependencies.
Why AI Still Can't Solve Your Real Mathematical Optimization Problem — The gap between LLM pattern-matching and rigorous combinatorial optimization remains wide, and ORPilot's hybrid approach points toward how specialized solvers will continue to outperform general models on constrained problems.
A Coding Guide to Implement a pgvector-Powered Semantic, Hybrid, Sparse, and Quantized Vector Search System — Using PostgreSQL as a full-featured vector database via pgvector reduces architectural complexity for teams that want semantic search without adding a dedicated vector store.
Tweaking Local Language Model Settings with Ollama — A deep dive into Ollama's configuration engine gives practitioners fine-grained control over local inference behavior without needing to modify model weights.
sqlite AGENTS.md — SQLite's addition of an AGENTS.md file — setting ground rules for AI agents pointed at its codebase — is a small but telling sign that major open-source projects are now actively managing AI agent interactions with their repos.

Watch This Week

Anthropic IPO timeline: With the Series H closed and run-rate revenue at $47B, watch for any formal S-1 filing signals or exchange selection announcements that would confirm 2026 as the IPO year.
AI agent security: The jqwik prompt injection incident is likely to trigger responses from package registries and AI coding tool vendors — watch for new scanning policies or sandboxing announcements from GitHub, npm, and similar platforms.
Waymo Ojai rollout: Public launch of Chinese-manufactured robotaxis in California and Arizona will test both regulatory tolerance and consumer acceptance, and could draw Congressional scrutiny given current US-China tech tensions.