Anthropic’s Claude Tag Is Learning Your Company, One Slack Message at a Time — TechCrunch

Anthropic today launched Claude Tag in research preview for Slack — a persistent AI teammate summoned via @Claude that maintains a shared identity across an entire organization and proactively surfaces information without being asked. Unlike point-in-time chatbot integrations, Claude Tag lets employees hand off half-finished tasks to each other via the same AI, accumulating institutional knowledge as it goes. The release, backed by a new agent identity access model, is a deliberate enterprise land-grab as Anthropic heads toward its anticipated IPO.

🤖 Frontier Models & Products

Alibaba’s AI Video Model Rises to No. 2 in Global Rankings — VentureBeat

Alibaba’s HappyHorse 1.1 has vaulted to second place in global AI video generation rankings, pushing past OpenAI’s Sora and ByteDance’s Seedance. The model is now live on Alibaba Cloud Model Studio with API access for enterprise integration, covering text-to-video, image-to-video, subject-to-video, and post-production editing. The jump underscores the narrowing gap between Chinese and US frontier video AI.

GLM-5.2 Raises the Bar for Open Models — TheZvi

TheZvi’s detailed analysis positions GLM-5.2 as the strongest openly available model to date, with benchmarks that substantially close the gap to leading closed systems while still trailing the very top frontier models. A recommended read for teams evaluating open-source alternatives to commercial APIs.

Anthropic Prepares Cowork Support for Mobile Apps — Testing Catalog

Anthropic plans to expand its Cowork task-scheduling system to mobile devices, letting users view and hand off Claude agent tasks across platforms.

claude-sonnet-5 Slug Spotted on Partner Provider — @synthwavedd on X

The model identifier claude-sonnet-5 has appeared on an Anthropic partner inference provider, suggesting a next-generation Sonnet is in late-stage preparation.

🔐 Security & Safety

OpenAI Launches GPT-5.5-Cyber and ‘Patch the Planet’ to Fix Open-Source Vulnerabilities at Scale — TechCrunch

OpenAI expanded its Daybreak cyber defense platform with GPT-5.5-Cyber — a specialized model scoring 85.6% on CyberGym, the highest single-model score recorded — and Patch the Planet, an open-source remediation initiative co-founded with Trail of Bits and partnered with HackerOne. More than 30 major projects including cURL, Go, Python, and Sigstore have committed to participate. The initiative’s framing is deliberate: AI has made finding bugs easy; automated patching at scale is now the bottleneck.

Anthropic Says Claude May Want to See Your ID — TechCrunch

Starting July 8, Anthropic may require government-issued ID verification (via Persona) for a small subset of flagged accounts — the company has not disclosed what triggers the check.

The Text in Claude Code’s ‘Extended Thinking’ Output Is Not Authentic — Patrick McCanna

Claude Code’s Extended Thinking reasoning is encrypted; users receive an AI-generated summary rather than the actual chain of thought, with full reasoning access gated to enterprise agreements.

Meta Pauses Employee Tracking Program After Sensitive Data Leaked Internally — Engadget

Meta suspended its Model Capability Initiative after keystroke logs, private conversations, and performance reviews became visible company-wide. Meta says no evidence of external access.

Cloudflare, Chrome, Firefox, and Edge Back New Privacy-First Anti-Bot Protocol — The Next Web

The PACT protocol verifies whether web traffic is human-generated without tracking individual users, with all three major browser vendors aligned on the spec.

Nearly Half of LG Smart TV Apps Are Laced with Residential Proxy SDKs — Spur

💻 Developer Tools & Agents

The AI World Is Getting ‘Loopy’ — TechCrunch

Agentic loops — autonomous systems where swarms of agents work continuously in the background without direct human invocation — are rapidly becoming a mainstream architectural pattern. The article identifies the hard problems: reliable stopping conditions, context rot prevention, agent-friendly tool design, and independent verification mechanisms. Essential reading as the industry standardizes around loop-based AI.

Claude Code 2.1.186 Changelog — Claude Code

New claude mcp login <name> / claude mcp logout <name> commands authenticate MCP servers directly from the CLI with --no-browser SSH support; also adds status filtering in /workflows and a Skills section in the /plugin installed tab.

Using Codex for Long-Running Projects — OpenAI

OpenAI’s official guide on treating Codex as a persistent multi-session workspace, covering task decomposition and human-autonomy balance for extended engineering projects.

Knowledge Agents: Beat Frontier Models with Better Structure — Weighty Thoughts

Smaller models augmented with structured knowledge injection (embeddings, proprietary data, multi-pass search) can match or exceed frontier models on specialized tasks — a practical case for domain-specific agent architectures over scaling.

An Agent Capability Library — Sami Honkonen

🏗️ Infrastructure & Compute

SpaceX Signs $6.3B Computing Deal with Open-Source AI Startup Reflection — CNBC

SpaceX has signed an agreement worth up to $6.3 billion giving Reflection AI access to Project Colossus’s Nvidia GB300 supercomputer cluster for training open-source models. The deal — coming alongside SpaceX’s earlier $60B Cursor acquisition — establishes SpaceX as a significant third-party AI compute provider, competing for cloud AI workloads alongside Microsoft Azure and Google Cloud.

Why American Data Centers Can’t Plug In — Works in Progress

A deep read on grid interconnection queue backlogs — not hardware or capital — as the primary bottleneck for the US AI infrastructure buildout, now stretching years at major grid operators.

Nvidia’s AI Data Center Design Runs Hotter to Use Far Less Water — The Verge

Nvidia’s Rubin-generation fully liquid-cooled reference design claims to eliminate most water consumption from data center cooling, addressing a leading criticism of AI infrastructure expansion.

Nvidia Seeks to Make Humanoid AI Robots Safer Around Humans — Bloomberg

Nvidia opened a pre-certification safety testing lab for humanoid robot makers, tackling contact and force-control challenges far more complex than autonomous vehicle safety.

🏛️ Policy & Society

Why Corporate AI Super PACs Spent $27 Million on a Local Election — The Verge

AI-backed corporate PACs poured $27 million into New York’s 12th Congressional District primary, targeting a state legislator with a record of AI regulation skepticism. The extraordinary spend for a local race signals the AI industry is treating electoral politics as a direct lever on regulatory outcomes.

Trump Signs Executive Orders to Boost Quantum Computing — WSJ

Two executive orders aim to accelerate US quantum computing development while establishing mitigation frameworks for quantum’s threats to existing cryptographic infrastructure.

Something’s Off with Midjourney’s Pivot to Body Scanners — The Verge

Midjourney announced an ultrasound body scanner claiming MRI-level diagnostic capability; The Verge finds the scientific evidence thin and the regulatory pathway undefined.

AI Is Cursing Renters with the Promise of Impossible Homes — The Verge

Generated by claude-sonnet-4-6 on 2026-06-23T10:00:00Z