GLM-5.2 boosts open models & Claude rumors and policy shifts - AI News (Jun 24, 2026)

Please support this podcast by checking out our sponsors: - Lindy is your ultimate AI assistant that proactively manages your inbox - https://try.lindy.ai/tad - Effortless AI design for presentations, websites, and more with Gamma - https://try.gamma.app/tad - Discover the Future of AI Audio with ElevenLabs - https://try.elevenlabs.io/tad Support The Automated Daily directly: Buy me a coffee: https://buymeacoffee.com/theautomateddaily Today's topics: GLM-5.2 boosts open models - Z.ai’s GLM-5.2 posts near-frontier benchmark results for an open-weights model, raising expectations for open AI and China lab releases. Keywords: GLM-5.2, open weights, benchmarks, distillation, coding. Claude rumors and policy shifts - Signals around Anthropic include an unconfirmed “claude-sonnet-5” identifier, upcoming Cowork mobile support, new ID checks, and debate over encrypted “thinking” logs. Keywords: Claude, Sonnet, Cowork, privacy, auditability. AI bubble fears and subsidies - Markets sold off as investors questioned AI valuations, while reporting argues big AI platforms have been subsidizing usage and may be forced into higher prices. Keywords: Nasdaq, valuations, margins, token billing, bubble. How big frontier models can get - A LessWrong analysis suggests inference speed, training compute, and data limits will shape model scaling through 2031, shifting constraints over time. Keywords: HBM bandwidth, inference, pretraining FLOPs, data ceiling, sparsity. SpaceX sells scarce GPU compute - SpaceX is turning its Colossus 2 data center into a compute business, with a major GB300 deal highlighting how GPU access has become a strategic choke point. Keywords: SpaceX, Colossus 2, Nvidia, GB300, compute contracts. Enterprise AI video reshuffles - Alibaba’s HappyHorse 1.1 aims at production AI video as rivals pull back, but geopolitical risk could complicate Western enterprise adoption. Keywords: AI video, HappyHorse, benchmarks, enterprise, compliance. Cheaper image inpainting breakthrough - Moebius claims near-industrial image inpainting quality at a fraction of the compute, pointing to a future where strong visual editing runs on smaller hardware. Keywords: inpainting, distillation, diffusion, lightweight models, edge. Agents move into super-apps - Tencent is testing an AI assistant inside Weixin, reflecting the broader shift from chatbots to agents that can take actions inside dominant apps. Keywords: Weixin, WeChat, Xiaowei, agents, mini-programs. - GLM-5.2 Debuts as Top Open-Weights Model, Still Behind Frontier - Rumored "Claude Sonnet 5" Identifier Spotted on Anthropic Partner Provider - Cory Doctorow argues the AI boom is a bubble built on hype and labor displacement - Analysis Projects Feasible Frontier Model Sizes Through 2031 Based on HBM, Serving Speed, Compute, and Data Limits - Anthropic tests Cowork integration in Claude iOS app, hinting at cloud-run tasks - Fastino’s Pioneer AI Page Highlights API Focus and Liability Disclaimers - Leaked Memo Reveals Meta’s AI Reorg Fueled Confusion and Record-Low Morale - OpenAI whitepaper explains how to use Codex for long-running projects - Alibaba launches HappyHorse 1.1 as AI video market reshuffles after Sora shutdown and Seedance pause - Claude Code ‘Extended Thinking’ Logs Are Encrypted Signatures, Not Full Reasoning - AI-Led Tech Sell-Off Hits US Stocks and Spreads Across Asian Markets - Moebius claims 10B-level image inpainting quality with a 0.22B-parameter model - OpenAI expands Daybreak with Codex Security upgrades, limited GPT-5.5-Cyber release, and open-source patching push - AI Platforms Face Backlash as Token Pricing Exposes Massive Usage Subsidies - Tencent Begins Small-Scale Test of Xiaowei AI Assistant Inside WeChat’s China App - Mercury launches Command, an AI assistant to run banking and finance workflows - Anthropic privacy policy allows ID and selfie checks for some Claude users - Structured “Knowledge Agents” Aim to Let Smaller Models Rival Frontier LLMs - SpaceX inks up-to-$6.3B Colossus compute deal with open-source AI startup Reflection Episode Transcript GLM-5.2 boosts open models Let’s start with open models, because there’s a genuine shake-up. Z.ai has released GLM-5.2, and early public evaluations paint it as a clear step up from GLM-5.1—possibly the strongest openly available model right now. Across several benchmark suites and coding leaderboards, it’s showing up surprisingly close to top closed models, sometimes in the neighborhood of Claude Opus. The big takeaway isn’t that benchmarks tell the whole story—users are split, with some calling it excellent for long-context coding and agent work, and others saying it feels “benchmaxxed,” overly verbose, or too eager to please. But it is a “sign of life” for open weights: the gap to the frontier looks narrower than it did a few months ago, even if it’s not closed. Claude rumors and policy shifts That open-weights theme also shows up in a different way: a new argument making the rounds that “knowledge agents” can narrow the gap without needing the biggest model. The idea is simple: wrap an LLM in a strong retrieval-and-structure harness—good indexing, sensible chunking, multiple retrieval passes—and you can get smaller or local models to perform much closer to frontier systems on specialized tasks. Why it matters: if pricing tightens, access changes, or policy removes capabilities, teams that can rely on structured private knowledge—not just raw model power—may be more resilient. AI bubble fears and subsidies Now, onto Anthropic, which had a busy day in the rumor mill and in policy changes. First, an unconfirmed but closely watched signal: someone claims the model identifier “claude-sonnet-5” appeared in a partner provider’s systems. That’s not an announcement, and it could mean a lot of things. Still, these backend slugs are often the earliest hint that an API update is nearing, and developers will be watching for capability or pricing shifts if a new Sonnet arrives. How big frontier models can get Second, more concrete: new UI elements spotted in a test build of the Claude iOS app suggest Anthropic is preparing mobile support for Cowork—its agent-style system for knowledge work. The language implies a shift toward cloud-executed tasks, which would be a meaningful usability jump. If you can schedule work from your phone without leaving a desktop session running, agents move from “cool demo” to something you can actually rely on day to day. SpaceX sells scarce GPU compute Third, and this is the one worth pausing on: a blogger digging through Claude Code’s local logs reports that “thinking blocks” aren’t readable reasoning at all—just a long signature. Anthropic’s documentation suggests the detailed reasoning is effectively sealed, and users only get a summarized version unless they have special access. The practical implication is about accountability: you can log inputs, outputs, and actions, but you may not be able to produce a verifiable chain-of-thought record after the fact. If you’re promising auditors, customers, or your own compliance team a full rationale trail, you’ll want to understand exactly what’s stored—and what isn’t. Enterprise AI video reshuffles Finally on Anthropic: a privacy policy update says some Claude users may be asked to verify age and identity by uploading a government ID and a selfie photo or video, via a third-party provider. Anthropic frames this as targeted—aimed at suspected fraud, with an appeal path. Even so, it raises the stakes: identity checks and biometric-adjacent verification add a new layer of data sensitivity, retention questions, and breach risk that users and enterprises will have to weigh. Cheaper image inpainting breakthrough Zooming out to the business side, markets reminded everyone how fast AI sentiment can flip. US tech and AI-linked shares sold off sharply, with the Nasdaq down and ripple effects across Asia—especially in chip-heavy markets. The catalyst wasn’t a single headline so much as a growing worry that valuations have sprinted ahead of sustainable profits, made worse by the prospect of higher interest rates. When money gets more expensive, “growth at any cost” becomes a tougher story to sell. Agents move into super-apps That connects to another uncomfortable narrative: the claim that major AI platforms have been subsidizing usage—offering far more compute than they’re charging for—hoping to raise prices later once customers are embedded. Reporting points to negative margins for heavy users and rising pressure to move from flat plans to stricter token billing and limits. If that trend accelerates, the impact is immediate: enterprise pilots get re-priced, CFOs demand clearer ROI, and teams start exploring smaller models, on-prem options, or retrieval-heavy setups that reduce token burn. Story 9 Cory Doctorow added a sharp social lens in a new interview, arguing today’s AI boom is being driven as much by financial and managerial incentives as by durable products. His framing is memorable: “centaurs” are workers using AI as a tool under their control, while “reverse centaurs” are workers turned into human appendages for automated systems—carrying blame when AI fails. Whether you agree or not, it’s a useful checkpoint: the productivity story isn’t just model quality, it’s who has agency, who has accountability, and whether organizations use AI to empower staff or to thin them out. Story 10 On the research-and-roadmap side, a LessWrong analysis tried to answer a question everyone asks but few quantify: how big can frontier models realistically get between 2023 and 2031? The headline is that inference limits—especially memory bandwidth and latency—cap what you can serve at decent speed, even if you can theoretically train something larger. Over time, the bottleneck shifts: serving feasibility loosens as hardware improves, but training compute and eventually unique data become the harder constraints. The significance is strategic: beyond a certain point, progress may come less from brute-size scaling and more from efficiency—sparsity, better data, better post-training, and better product integration. Story 11 Speaking of hardware, SpaceX is increasingly acting like a compute provider. It’s signed a major agreement with open-source AI startup Reflection AI for access to top-end Nvidia GB300 chips at its Colossus 2 data center. The key point isn’t just the eye-watering spend—it’s what it signals: scarce GPU access is now a competitive moat, and anyone who can warehouse and allocate cutting-edge capacity can sell it like a utility. The AI race is starting to look as much like supply chain and infrastructure as it does like algorithms. Story 12 In generative media, Alibaba Cloud released HappyHorse 1.1, positioning it as an enterprise-ready AI video model at a moment when the field has been reshuffling. With some rivals pulling back due to costs or copyright friction, Alibaba is leaning on strong benchmark placement and global cloud distribution to win business workflows. The catch is geopolitical: procurement scrutiny is rising for Chinese providers in Western markets, and for regulated buyers, “best model” isn’t enough—you also need a clear story on compliance, residency, and risk. Story 13 And for image editing, researchers introduced Moebius, a lightweight inpainting system that claims quality comparable to much larger models while dramatically cutting compute. If these results hold up in broader use, it’s part of a bigger pattern: not every improvement comes from bigger models. Sometimes the most enabling breakthroughs are the ones that make strong capabilities cheap and portable—meaning more tools can run on modest GPUs, or closer to the edge. Story 14 Lastly, the agent trend keeps moving into the apps people already live in. Tencent says it’s testing an AI assistant called Xiaowei inside Weixin, with the ability to act inside the super-app—like helping with messaging or launching mini-programs. That “embedded agent” distribution advantage is hard to overstate when you’re sitting on more than a billion users. Meanwhile, OpenAI published guidance on using Codex across long-running, multi-session work—treating it more like a persistent workspace than a one-and-done prompt tool. In practice, that’s where real productivity either happens or falls apart: continuity, verification, and handoffs. Story 15 And on security, OpenAI’s Daybreak initiative is pushing beyond vulnerability finding toward automating fixes—scanning code, validating issues, and generating patches that fit into existing security workflows. The important shift is operational: AI security value increasingly depends on governance, review, and integration, not just raw model scores. Subscribe to edition specific feeds: - Space news * Apple Podcast English * Spotify English * RSS English Spanish French - Top news * Apple Podcast English Spanish French * Spotify English Spanish French * RSS English Spanish French - Tech news * Apple Podcast English Spanish French * Spotify English Spanish Spanish * RSS English Spanish French - Hacker news * Apple Podcast English Spanish French * Spotify English Spanish French * RSS English Spanish French - AI news * Apple Podcast English Spanish French * Spotify English Spanish French * RSS English Spanish French Visit our website at https://theautomateddaily.com/ Send feedback to [email protected] Youtube LinkedIn X (Twitter)