AI Radar — 13 Jul 2026
AI Radar — 13 Jul 2026
Cursor ships side chats and iOS agents; Apple files trade secret suit against OpenAI; Meta’s in-house chip enters TSMC production; Claude Code enables auto mode by default on cloud providers.
Run: Jul 10–13 2026 (strict 72h window) · 36 reviewed → 4 published · 2 verified · 2 secondary · 0 rumor · 36% exploration
Note: The Jul 10 bulletin covered items from the Jul 7–10 window (GPT-5.6 Sol/Terra/Luna, ChatGPT Work, Grok 4.5, VS Code 1.128, Claude Reflect, Kimi K2.7, Meta Muse Image, Automox MCP 2.2). Those are not re-published here. This bulletin covers the quieter Jul 10–13 follow-on window.
TL;DR
- Cursor 3.11 — side chats, agent transcript search, and team MCP distribution ship July 10; iOS public beta opens to all paid plans. (→ Cursor 3.11)
- Apple sues OpenAI — trade secret theft complaint filed July 10 in Northern California federal court, naming OpenAI’s hardware chief as defendant; lawsuit clouds OpenAI’s IPO runway. (→ Apple v. OpenAI)
- Meta Iris chip — Meta’s first custom AI accelerator enters mass production with TSMC in September, targeting 14 gigawatts of compute by 2027. (→ Meta Iris)
- Claude Code 2.1.207 — auto mode activates by default on Bedrock, Vertex AI, and Foundry deployments; Bedrock defaults to Opus 4.8. (→ Claude Code 2.1.207)
Items
Cursor 3.11 Ships Side Chats, Transcript Search, and Team MCP Distribution
Source: Cursor — https://cursor.com/changelog · Jul 10 2026 · T2 · verified Verification: T2 verified · changelog · dev-tools · workflow-automation
Cursor 3.11 released July 10, introducing side chats that run in parallel to the main agent conversation — each is a full, durable agent session opened via /side, /btw, or the panel plus button, with context inherited from the primary thread. Agent transcript search, triggered from the command palette, builds a local index over all historical conversations and scales to thousands of sessions. Team administrators can now configure MCP servers once in the Dashboard and distribute them to cloud agents, the agent window, the IDE, and the CLI simultaneously; team members install approved integrations from a shared marketplace without writing individual configs. Cursor for iOS entered public beta on all paid plans on the same date, enabling developers to launch cloud agents, review diffs, and approve pull requests from a mobile device.
Why it matters for automation/productivity: Team MCP server distribution eliminates per-developer configuration overhead for integrations connecting coding agents to existing tools — a concrete lever for organizations standardizing agentic pipelines at team scale. The iOS public beta enables asynchronous task management: cloud agents run in isolated VMs while the developer reviews progress or triggers the next step from a phone, without requiring a desktop session to stay active.
Key claims:
- Side chats, transcript search, redesigned pickers → cursor.com/changelog (Jul 10 2026, T2 primary verified)
- Team MCP server distribution, team marketplace → cursor.com/changelog
- iOS public beta on all paid plans → cursor.com/changelog; cursor.com/blog/ios-mobile-app (Jun 29 2026, T2 blog, confirms prior TestFlight was invite-only)
Caveats: Transcript search requires a locally built index; performance on air-gapped or very low-disk environments is untested. iOS features at launch exclude direct computer-use capabilities present on the desktop app.
Cross-references:
- https://cursor.com/blog/ios-mobile-app (T2, background on iOS remote control and cloud agent handoff)
Apple Files Trade Secret Lawsuit Against OpenAI
Source: CNBC — https://www.cnbc.com/2026/07/10/apple-openai-lawsuit-trade-secrets.html · Jul 10 2026 · T2; CNN Business — https://www.cnn.com/2026/07/10/tech/apple-openai-devices-lawsuit · Jul 10 2026 · T2; Semafor — https://www.semafor.com/article/07/10/2026/apple-sues-open-ai-accusing-it-of-stealing-trade-secrets · Jul 10 2026 · T2 Verification: T2 secondary · announcement · policy-regulation · ai-for-business
Apple filed suit in the Northern California federal court on July 10, naming OpenAI and two former Apple engineers as defendants: Tang Tan, now OpenAI’s chief hardware officer and previously Apple’s VP of product design who led iPhone, Apple Watch, AirPods, and iPod development; and Chang Liu, a former senior systems electrical engineer. The complaint alleges that Tan directed candidates still employed at Apple to bring hardware components and documentation to OpenAI interviews for show-and-tell sessions, and that Liu used an Apple-issued laptop to download technical documents before joining OpenAI. Apple is seeking damages, injunctions, and an order compelling OpenAI to stop using the alleged trade secrets. OpenAI responded: “We have no interest in other companies’ trade secrets.”
Why it matters for automation/productivity: The case introduces legal uncertainty into OpenAI’s IPO timeline — analysts had expected a historically large offering in 2026, and an unresolved federal lawsuit is a material disclosure risk. Teams building enterprise AI workflows on OpenAI’s platform should factor this into vendor risk assessments alongside pricing and availability. More broadly, the case may set precedent affecting how AI companies conduct hardware-engineer recruiting from device manufacturers.
Key claims:
- Filed July 10 in Northern California federal court → CNBC, CNN Business (both Jul 10 2026, T2)
- Tang Tan as defendant (former Apple VP, now OpenAI hardware chief) → CNBC, Semafor
- Show-and-tell interview pattern allegation; Liu device-download allegation → CNBC, CNN Business
- OpenAI denial: “We have no interest in other companies’ trade secrets.” → multiple outlets (Jul 10 2026)
Caveats: The complaint is civil, not criminal. Trade secret cases in tech frequently settle before trial. The filing represents Apple’s allegations; an independent finding of liability has not been made. Bloomberg’s Jul 11 coverage notes that an OpenAI engineer’s internal message about the allegations was characterized by OpenAI as context-collapsed.
Conflict of interest note: Apple and OpenAI have had a commercial partnership (ChatGPT integration in iOS) since 2024; that relationship and the lawsuit are running simultaneously.
Cross-references:
- https://www.bloomberg.com/news/articles/2026-07-11/openai-engineer-s-lol-moment-set-stage-for-legal-fight-with-apple (T2, additional defendant context, Jul 11)
- https://www.gurufocus.com/news/8954632/openai-faces-legal-challenges-from-apple-amid-ipo-plans (T3, IPO impact framing)
Meta’s Iris AI Chip Enters Production Track for September
Source: Reuters via multiple news outlets · Jul 10 2026 · T2-T3 (original Reuters report not directly fetched due to HTTP 403); confirmed via ResultSense — https://www.resultsense.com/news/2026-07-10-meta-iris-ai-chip-production/ · Jul 10 2026 · T3; FullStackEvolved — https://www.fullstackevolved.com/blog/meta-iris-ai-chip-september-2026-07-10/ · Jul 10 2026 · T3 Verification: T2 secondary (Reuters sourcing confirmed via T3 outlets) · announcement · ai-for-business
Meta confirmed on July 10 that its first in-house AI accelerator, code-named Iris, is scheduled to enter mass production at TSMC in September 2026. The chip was designed with Broadcom and passed internal testing in approximately six weeks without major blockers. Meta plans to release updated Iris versions at roughly a six-month cadence through 2027 and expects to reach 14 gigawatts of compute capacity by 2027, double its current 7-gigawatt baseline. Meta’s projected AI infrastructure spend is up to $145 billion in 2026, with Iris part of a strategy to reduce reliance on Nvidia and AMD processors. The Iris announcement follows the July 1 Bloomberg report that Meta is building Meta Compute, a cloud service to sell excess GPU capacity to outside customers.
Why it matters for automation/productivity: Custom silicon at Meta scale accelerates the path toward compute commoditization. If Iris production succeeds and Meta Compute launches as planned, AI inference pricing pressure on Nvidia-dependent providers is likely — a structural signal for teams that model AI API cost trajectories. For Meta-native workflows (WhatsApp, Instagram, Messenger integrations), Iris will underpin the compute powering Meta AI features, which may improve latency and reduce rate-limiting constraints for business API users.
Key claims:
- September production start, TSMC manufacturing, Broadcom design → Reuters via multiple T3 outlets (Jul 10 2026); T2 corroboration via Gurufocus/CNBC citations
- Testing completed in approximately six weeks → ResultSense (Jul 10, citing Reuters)
- Target: 14 GW by 2027, from current 7 GW → FullStackEvolved (Jul 10)
- Six-month chip release cadence through 2027 → FullStackEvolved
- Up to $145B AI infrastructure spend in 2026 → CNBC via Gurufocus (T2)
Caveats: Reuters primary was not directly fetched (HTTP 403 during this run); all details sourced from T3 outlets citing Reuters. Production start date is Meta-stated; chip yield and performance claims will require independent confirmation after production. Iris is an ASIC optimized for recommendation ranking, not a general-purpose GPU — applicability to Llama training and large-context inference workloads will depend on second-generation designs.
Cross-references:
- https://www.gurufocus.com/news/8954665/meta-meta-stock-rises-after-announcement-of-inhouse-ai-chip-production (T3, market reaction; Meta stock up ~6% on the announcement)
- https://finance.yahoo.com/technology/ai/articles/meta-compute-launch-sends-ai-015258040.html (T3, Meta Compute context from Jul 1 announcement)
Claude Code 2.1.207 Activates Auto Mode by Default on Cloud Providers
Source: Claude Code changelog — https://code.claude.com/docs/en/changelog · Jul 11 2026 · T2 · verified; Releasebot — https://releasebot.io/updates/anthropic/claude-code · Jul 11 2026 · T3 (corroborating) Verification: T2 verified · changelog · dev-tools
Claude Code version 2.1.207, released July 11, activates auto mode by default on Bedrock, Vertex AI, and Foundry deployments — removing the previous CLAUDE_CODE_ENABLE_AUTO_MODE environment variable requirement. Teams that want to opt out can disable the feature through managed settings. Bedrock deployments now default to Claude Opus 4.8. The release also resolves terminal freezing and keystroke lag during streaming of long code blocks or large tables, fixes spurious prompt-injection warnings on benign system-generated updates, and blocks plugin hook commands containing shell-injection-vulnerable patterns.
Why it matters for automation/productivity: Auto mode default behavior directly affects enterprise teams running Claude Code headlessly in CI/CD pipelines on Bedrock, Vertex AI, and Foundry — new deployments now get agentic task completion without extra configuration steps. The shell injection fix in plugin hooks closes a security gap relevant for organizations that allow user-supplied plugin inputs in Claude Code workflows. The terminal lag fix is a quality-of-life improvement for developers using Claude Code on large codebases.
Key claims:
- Auto mode now default on Bedrock, Vertex AI, Foundry → Claude Code changelog (Jul 11 2026, T2 primary verified)
- Bedrock defaults to Opus 4.8 → changelog
- Terminal freeze fix, prompt injection false-positive fix, shell-injection block → changelog
Caveats: No migration guide was published alongside this release; teams with existing Bedrock/Vertex/Foundry deployments running in non-auto mode will need to explicitly set the opt-out flag to preserve prior behavior after upgrade.
Dropped
Items considered but not published, with reason.
| Title considered | Source | Reason |
|---|---|---|
| GPT-5.6 Sol/Terra/Luna GA | openai.com | Published in Jul 10 bulletin (Jul 9 event) |
| ChatGPT Work desktop agent | engadget.com | Published in Jul 10 bulletin (Jul 9 event) |
| Grok 4.5 coding and agentic model | axios.com | Published in Jul 10 bulletin (Jul 8 event) |
| VS Code 1.128 multi-chat agent sessions | code.visualstudio.com | Published in Jul 10 bulletin (Jul 8 event) |
| Claude Reflect usage analytics dashboard | anthropic.com | Published in Jul 10 bulletin (Jul 9 event) |
| Kimi K2.7 expands to Copilot Business/Enterprise | github.blog | Published in Jul 10 bulletin (Jul 7 event) |
| Meta Muse Image first in-house image model | cnbc.com | Published in Jul 10 bulletin (Jul 7 event) |
| Automox MCP Server 2.2 visual review | globenewswire.com | Published in Jul 10 bulletin (Jul 7 event) |
| SambaNova $1B Series F at $11B valuation | bloomberg.com / techcrunch.com | Jul 8 2026 — outside strict Jul 10–13 window |
| Fed Productivity & Jobs task force (Andreessen) | washingtonpost.com | Jul 9 2026 — outside strict window |
| Prime Intellect $130M Series A (Radical Ventures) | techcrunch.com | Jul 8 2026 — outside strict window |
| Anthropic Claude Cowork web/mobile expansion | anthropic.com | Jul 7 2026 — outside window; published in Jul 8 bulletin |
| Claude Desktop Linux beta (Ubuntu/Debian) | anthropic.com | Jun 30 2026 — outside window |
| OpenAI gpt-oss-120b / gpt-oss-20b open-weight models | openai.com | Dec 2025 — outside window; resurfaced in aggregator coverage |
| Anthropic $47B revenue run rate / Series H | anthropic.com / x.com | May 2026 — outside window; resurfaced in July roundups |
| Qualcomm acquires Modular ($3.92B all-stock) | qualcomm.com | Jun 24 2026 — outside window |
| NVIDIA Nemotron 3 Nano Omni open multimodal | nvidia.com | Apr 28 2026 — outside window |
| Mistral $830M data center debt financing | cnbc.com | Mar 30 2026 — outside window |
| Databricks $850M UK investment | databricks.com | Mar 31 2026 — outside window |
| NAVER + NVIDIA sovereign AI infrastructure | nvidianews.nvidia.com | Jun 7 2026 — outside window |
| Accenture Edge + Google Cloud mid-market agentic | newsroom.accenture.com | Jul 7 2026 — outside strict window |
| WAIC 2026 Shanghai July 17–20 pre-announcement | globaltimes.cn | Jul 7–8 2026 — outside strict window; conference starts Jul 17 |
| Gemini in Sheets expands to 11 more languages (incl. Indonesian) | workspaceupdates.googleblog.com | Jul 7 2026 — outside strict window |
| Snowflake CoWork Deep Research GA | snowflake.com | Jul 7 2026 — outside strict window |
| Gemini 3.5 Pro target date July 17 (leaked spec) | techtimes.com | Unconfirmed — leaked spec only, no Google primary confirming date or pricing |
| MCP Servers 2026.7.10 maintenance release | modelcontextprotocol/servers | Single T3 source only; unable to verify from github.com/modelcontextprotocol/servers primary during this run |
| Meta Compute GPU cloud service launch | bloomberg.com (Jul 1) | Announced Jul 1; no confirmed in-window launch date from primary source |
| AgentPrizm AgentMemory/AgentSkills platform | aiagentstore.ai | Jul 9 2026 — outside strict window; T3 source only |
| Microsoft 365 AI auto-attendant updates | techcommunity.microsoft.com | Low BD actionability — enterprise telephony feature, no agentic execution capability |
| Gemini Omni video generation global rollout | blog.google | Announced at Google I/O May 19 2026; API GA Jun 30; no new in-window announcement |
| Gemini Spark / Daily Brief Anthropic-comparable features | gemini.google | Google I/O announcements (May 2026); rollout continuation, not a new Jul 10–13 launch |
Limitations
- Sources unreachable: Reuters.com returned HTTP 403 during this run — Meta Iris chip production story sourced from T3 outlets citing Reuters rather than the Reuters primary. CNBC Apple lawsuit article returned HTTP 403; same story confirmed via CNN Business (T2, accessible) and Semafor (T2, accessible). Cursor changelog was accessible. Claude Code changelog accessible.
- Login-walled coverage: X/Twitter timelines, LinkedIn, Instagram, Discord, and Slack communities were not directly accessed. Public X posts visible via search engines were captured; no X-first announcements were identified in this window that required login access.
- Overlap with prior bulletins: This window (Jul 10–13) significantly overlaps with the Jul 10 bulletin’s period (Jul 7–10). Twenty-eight items from Jul 7–10 were reviewed and dropped here as previously published. Only four new items originating from Jul 10 or later met the quality bar.
- Quiet period characteristics: The Jul 10–13 window immediately follows a dense Jul 7–10 release cluster (GPT-5.6, ChatGPT Work, Grok 4.5, VS Code 1.128, Claude Reflect). Post-cluster windows typically yield fewer new launches; this window confirms that pattern.
- Categories with thin coverage: model-release — no new frontier or open-weight model launched Jul 10–13; Gemini 3.5 Pro target date (Jul 17) is leaked but not confirmed from Google primary. mcp-ecosystem — only the unverifiable MCP maintenance release; no significant MCP server launch confirmed. agent-framework — no standalone framework releases. research-papers — no high-relevance papers surfaced. workflow-automation — no new end-user autonomous agent product launched in the window.
- Items requiring upgrade: Meta Iris chip (T2 secondary) — upgrades to T2/verified once Reuters primary is directly accessible. Apple v. OpenAI (T2 secondary) — will upgrade to verified once the court filing is publicly accessible via PACER or direct Apple newsroom statement.
- Vendor benchmark caveats: No vendor benchmarks were cited in this bulletin. Meta $145B AI spend figure is Meta-stated guidance; actual Q2 2026 spend pending earnings.
- Geographic bias: Coverage is US-heavy. Targeted SEA-region search surfaced no in-window AI product launches or policy developments from Indonesia or broader Southeast Asia. A Gemini Sheets language expansion to Indonesian was noted but is outside the strict window (Jul 7). SEA coverage gap is structural and documented across prior runs.
- Time period bias: This bulletin covers Thursday–Sunday. Items from Jul 13 specifically (the final day of the window) are sparse — search indexing for late-in-day announcements on a weekend lags.
Search log (compact)
Q: "Anthropic Claude announcement July 2026" → 10 results, 3 high-relevance
Q: "OpenAI announcement release July 11 12 13 2026" → 8 results, 3 high-relevance
Q: "Google DeepMind Gemini release July 2026" → 9 results, 2 high-relevance
Q: "xAI Grok new model release July 2026" → 10 results, 2 high-relevance (covered prior bulletin)
Q: "Meta AI announcement July 11 12 13 2026" → 10 results, 3 high-relevance
Q: "Anthropic Claude Fable 5 restored redeployed July 2026" → 9 results, 2 high-relevance (Jul 1 — outside window)
Q: "Gemini 3.5 Flash release generally available July 2026" → 8 results, 1 high-relevance (May launch — outside window)
Q: "AI agent framework launch release July 11 12 13 2026" → 10 results, 2 high-relevance [exploratory]
Q: "MCP Model Context Protocol server release July 2026" → 10 results, 1 high-relevance
Q: "Meta Compute cloud AI chip Iris July 2026" → 10 results, 7 high-relevance
Q: "Anthropic revenue ARR $47 billion overtake OpenAI July 2026" → 8 results, 2 high-relevance (May event — outside window) [exploratory]
Q: "OpenAI open source model gpt-oss-120b gpt-oss-20b release 2026" → 9 results, 1 high-relevance (Dec 2025 — outside window) [exploratory]
Q: "Apple lawsuit OpenAI trade secret employees July 11 2026" → 10 results, 9 high-relevance
Q: "Cursor AI IDE update release July 2026" → 8 results, 6 high-relevance
Q: "LangChain LlamaIndex CrewAI update release July 10 11 12 2026" → 9 results, 1 high-relevance
Q: "Prime Intellect $130 million Series A funding date July 2026" → 9 results, 2 high-relevance (Jul 8 — outside window)
Q: "Anthropic Claude new feature announcement July 11 12 13 2026" → 9 results, 4 high-relevance
Q: "Google Gemini product update July 11 12 2026" → 9 results, 2 high-relevance (rollout of May announcements)
Q: "Meta Compute official launch date July 10 11 2026" → 9 results, 3 high-relevance (Jul 1 announced — outside window)
Q: "AI news July 10 11 12 13 2026 new launch announcement" → 10 results, 5 high-relevance [exploratory]
Q: "OpenAI GPT-5.6 public release all users July 11 2026" → 9 results, 4 high-relevance (Jul 9 event — prior bulletin)
Q: "Anthropic Claude Code update release July 10 11 12 2026" → 10 results, 6 high-relevance
Q: "site:x.com AI announcement July 11 12 2026" → 9 results, 3 high-relevance [social search]
Q: "new AI model open source release July 10 11 12 2026" → 9 results, 2 high-relevance
Q: "AI Indonesia Asia Tenggara startup AI July 2026" → 8 results, 0 high-relevance [geo-diversity]
Q: "GitHub trending AI repositories July 2026 week" → 8 results, 2 high-relevance [exploratory Stage 3.5]
Q: "buildfastwithai July 12 OR July 11 2026 AI news biggest stories" → 9 results, 5 high-relevance [exploratory]
Q: "Cursor 3.11 problems limitations criticism July 2026" → 8 results, 1 high-relevance [adversarial]
Q: "Apple OpenAI lawsuit skepticism criticism IPO impact July 2026" → 8 results, 4 high-relevance [adversarial]
Q: "AI news July 13 2026 latest announcements model release" → 9 results, 2 high-relevance [exploratory]
Q: "Claude Desktop Linux beta release date 2026" → 9 results, 1 high-relevance (Jun 30 — outside window)
Q: "Qualcomm acquire Modular $3.92 billion announcement date July 2026" → 10 results, 1 high-relevance (Jun 24 — outside window)
Q: "NVIDIA Nemotron 3 Nano Omni release date July 2026" → 9 results, 1 high-relevance (Apr 28 — outside window)
Q: "Mistral AI $830 million debt financing data center July 2026 date" → 9 results, 1 high-relevance (Mar 30 — outside window)
Q: "Databricks UK investment $850 million date July 2026" → 9 results, 1 high-relevance (Mar 31 — outside window)
Q: "NAVER NVIDIA sovereign AI infrastructure date July 2026" → 9 results, 1 high-relevance (Jun 7 — outside window)
Q: "Federal Reserve AI economic advisory board Andreessen July 11 2026" → 9 results, 3 high-relevance (Jul 9 — outside window) [exploratory]
Q: "SambaNova $1 billion Series F funding date July 2026" → 9 results, 2 high-relevance (Jul 8 — outside window)
Q: "MCP ecosystem new server workflow automation July 10 11 12 2026" → 8 results, 2 high-relevance [exploratory Stage 3.5]
Q: "new AI coding tool developer productivity launch July 12 13 2026" → 8 results, 1 high-relevance
Q: "Hugging Face daily papers trending July 11 12 2026" → 7 results, 1 high-relevance [exploratory Stage 3.5]
Q: "Gemini Omni video generation rollout date July 2026" → 8 results, 1 high-relevance (May/Jun launch — outside window)
Q: "Apple OpenAI consequences developer talent mobility criticism July 2026" → 8 results, 5 high-relevance [adversarial]
Q: "AI product launch announcement July 11 12 13 2026 new tool" → 9 results, 2 high-relevance [exploratory]
Q: "WAIC World AI Conference Shanghai July 17 2026 announcements" → 8 results, 1 high-relevance (conference Jul 17 — outside window) [exploratory]
Q: "Databricks UK investment $850 million announcement date July 2026" → 9 results, 1 high-relevance (Mar 31 — outside window)
Q: "Snowflake AI product announcement July 2026" → 8 results, 1 high-relevance (Jul 7 — outside window)
Q: ""AI startup" funding announcement July 10 11 12 2026" → 8 results, 1 high-relevance [exploratory]
Q: "Accenture Google Cloud AI July 2026 partnership" → 9 results, 1 high-relevance (Jul 7 — outside window)
Q: "Meta Compute cloud service launch date July 2026" → 10 results, 3 high-relevance
Q: "Cursor 3.11 iOS mobile beta Team MCP servers July 10 2026 primary" → 10 results, 7 high-relevance
Q: "AI agent OR autonomous AI launch announcement July 11 12 13 2026" → 9 results, 1 high-relevance [exploratory]
Q: "GPT-5.6 Apple OpenAI lawsuit criticism controversy July 2026" → 7 results, 4 high-relevance [adversarial]
Q: "AI research paper arXiv Hugging Face notable July 10 11 12 2026" → 7 results, 1 high-relevance [exploratory Stage 3.5]
Q: "Microsoft AI announcement July 10 11 12 2026" → 9 results, 1 high-relevance
Q: "new AI model release July 12 13 2026 launch" → 8 results, 1 high-relevanceTotal searches: 56, of which 20 exploratory or adversarial (36%).
Suggested next runs
- Gemini 3.5 Pro launch July 17 — leaked spec places GA on July 17 with a 2M-token context window; watch ai.google.dev for the model card and confirmed API pricing before committing to benchmarks
- Apple v. OpenAI case progress — PACER docket watch for OpenAI motion to dismiss, which would be the first filing indicating how the company intends to fight the trade secret allegations
- Meta Iris production results — September production start; first third-party benchmarks on Iris performance (inference latency, power efficiency vs. H100/H200) will surface in Q4 earnings or Meta AI blog
- MCP 2026-07-28 spec final publish — RC locked May 21; final specification ships July 28; SDK Tier 1 implementers expected to ship support within the ten-week window; watch blog.modelcontextprotocol.io for breaking changes to stateless core and OAuth alignment