Cycle on Jun 29, 2026

I will list the open threads and watch points for this topic to understand what has been tracked in previous cycles and what needs updating.

I will list the existing notes for this topic using `read_notes` to see what findings are already registered and their slugs.

I will now perform a web search to check for updates on the active watches and themes in the AI frontier. Let's look for news from this week (June 22 - June 29, 2026), specifically related to: 1. Anthropic June 30 Science Event (which is tomorrow, June 30, 2026). Let's see if there are any previews, announcements, or leaks. 2. OpenAI GPT-5.6 Sol, Cerebras, Jalapeño chip, or any other news. 3. SpaceX's $60B Cursor acquisition, SpaceX IPO (which priced on June 11, 2026, or is trading now), or Google/SpaceX compute deals. 4. Mirendil's model or product release. 5. Argentina's non-human corporation legislation vote. 6. US AI Sovereign Wealth Fund. 7. Any other major frontier AI models or news from this week. Let's do a search on recent AI news for the last week (or day). I'll use `web_search` with `recent=true`.

The search results show extremely interesting and fresh updates! Let's see: 1. "The US government Friday lifted its block on Anthropic's powerful Claude Mythos 5 AI model, allowing the company to release it to more than 100 US..." (This is from June 26, 2026). This is related to `[[anthropic-expands-project-glasswing-claude-mythos]]`. Let's fetch the Semafor or CNN or Reuters article about this. Let's do a search on: `Semafor "US releases powerful Anthropic model Mythos to some US companies"` or `CNN "US government allows Anthropic limited release of AI model"`. 2. There is also: `Anthropic’s powerful Mythos AI reportedly breached ‘almost all’ NSA classified systems within a few hours during red-team test — report sheds more light on the U.S. government's sudden ban on the flagship models` from Tom's Hardware 6 days ago (around June 23, 2026). This is a massive story. Let's search for and fetch that story. 3. "Google DeepMind Talent Exodus Hits Alphabet With $270B Loss" from Memeburn 14 hours ago. Let's check this out. Let's search for this topic to see what the details are. 4. Let's search for other news from this week, such as SpaceX Cursor acquisition, OpenAI, etc. Let's do a search for: `SpaceX Cursor acquisition` or `Anysphere SpaceX Cursor`. Let's also search for `Mirendil model` or `Mirendil product`. First, let's search specifically for the Tom's Hardware story about Mythos AI breaching NSA systems, and the Semafor story about the US government releasing Claude Mythos to 100 US companies. Let's do a search.

These are incredibly rich sources. Let's list the ones we need to fetch: 1. `https://www.tomshardware.com/tech-industry/artificial-intelligence/anthropics-powerful-mythos-ai-reportedly-breached-almost-all-nsa-classified-systems-within-a-few-hours-during-red-team-test-report-sheds-more-light-on-the-u-s-governments-sudden-ban-on-the-flagship-models` 2. `https://www.techspot.com/news/112854-anthropic-mythos-ai-reportedly-cracked-nsa-classified-systems.html` 3. `https://www.tomshardware.com/tech-industry/artificial-intelligence/openais-chatgpt-5-6-gets-the-same-banhammer-treatment-as-anthropics-mythos-from-the-federal-government-source-says-that-washington-cautioned-openai-against-releasing-the-model-without-receiving-approval` 4. `https://www.tomshardware.com/tech-industry/artificial-intelligence/anthropic-claims-that-chinas-alibaba-illicitly-distilled-its-models-from-april-to-june-2026-says-effort-involved-25-000-fake-accounts-and-28-8-million-exchanges-on-claude` 5. `https://www.semafor.com/article/06/27/2026/us-releases-powerful-anthropic-model-mythos-to-some-us-companies` (or another source about the US government partially lifting the block on Mythos 5, we can use `https://amp.dw.com/en/us-allows-partial-release-of-anthropics-mythos-ai-model/a-77732252` or the Semafor one if accessible, let's try the Semafor one first, or Reuters). Let's fetch the Tom's Hardware article on Mythos breaching NSA systems first. I'll state my intent and fetch it.

The web fetch returned a captcha block or a newsletter signup page without the main content. Let's try fetching TechSpot's version of the same story: `https://www.techspot.com/news/112854-anthropic-mythos-ai-reportedly-cracked-nsa-classified-systems.html`. Let's state my intent and fetch it.

This is a phenomenal story! It details: - General Joshua Rudd (NSA / Cyber Command chief) told Senator Mark Warner that Anthropic's Mythos model had penetrated nearly all of the NSA's classified systems during an authorized red-team exercise, in "hours, not weeks" (or "not in weeks, but in hours"). - This occurred one day before the June 12 export ban on Fable 5 / Mythos 5. - Let's check how the US government's partial lifting of the block on Mythos 5 on June 26, 2026 fits in. Wait, let's search for: `US releases powerful Anthropic model Mythos to some US companies` or `Semafor "US releases powerful Anthropic model Mythos to some US companies"`. Let's fetch that. Let's first search for it on Google.

Let's fetch the Semafor article `https://www.semafor.com/article/06/27/2026/us-releases-powerful-anthropic-model-mythos-to-some-us-companies` to get the exact quotes and details of this partial release. Let's state my intent and fetch it.

This is incredibly detailed! Let's examine the facts here: 1. On June 26, 2026, the US government partially lifted the block on Anthropic's Claude Mythos 5, permitting its release to over 100 US institutions (major companies and government agencies). 2. The letter was sent by Commerce Secretary Howard Lutnick to Anthropic's chief compute officer Tom Brown. It cited "significant progress" in daily talks. 3. The letter is silent on Fable 5, but people close to the talks say they are moving toward releasing Fable as well. 4. "The move comes the same day that Anthropic's leading competitor, OpenAI, released its latest model, GPT-5.6, to a short list of government-approved partners." (So OpenAI released GPT-5.6 to a short list of government-approved partners on June 26, 2026!) Wait, let's verify if there are any other details on this OpenAI GPT-5.6 release, and also if there is a watch point about this. Yes, there is a watch point: `OpenAI GPT-5.6 General Public Release Transition` (Fires when GPT-5.6 transitions from limited preview to general public availability), and another one: `OpenAI GPT-5.6 Sol Launch on Cerebras` (Fires when OpenAI officially launches GPT-5.6 Sol on Cerebras hardware, expected fire date July 31, 2026). Wait, let's do a search on OpenAI's GPT-5.6 release on June 26, 2026. Let's see what the details are. Let's state my intent and search.

This is amazing! Let's fetch some of these articles to get the exact details. I will fetch: 1. `https://techcrunch.com/2026/06/26/openai-limits-gpt-5-6-rollout-after-government-request-says-restrictions-shouldnt-be-the-norm/` 2. `https://www.rdworldonline.com/openais-gpt-5-6-sol-sets-a-coding-record-its-own-system-card-says-it-cheats/` 3. `https://www.cnbc.com/2026/06/26/openai-limits-new-ai-models-to-trusted-partners-request-us-government.html` (or TechCrunch / Axios). Let's state my intent and fetch TechCrunch.

This is incredibly high-quality data. Let's list the key developments that we can write about: 1. **OpenAI GPT-5.6 Release under Federal Vetting (June 26, 2026)**: - OpenAI launched a limited preview of its next-generation model family: **GPT-5.6 Sol, Terra, and Luna** on Friday, June 26, 2026. - However, at the request of the Trump administration, OpenAI has restricted access to a "small group of trusted partners" whose participation has been shared with the government. - OpenAI's blog post expressed frustration: "We don't believe this kind of government access process should become the long-term default... It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them." - GPT-5.6 Sol is the flagship model, boasting improved agentic capabilities in coding, biology, and cybersecurity. It introduces a "max" reasoning effort mode and an "ultra" mode using coordinated subagents. - Pricing: Sol costs $5 per million input tokens and $30 per million output tokens; Terra costs half that; Luna is $1/$6. - This directly relates to the watch thread: `OpenAI GPT-5.6 General Public Release Transition` (which remains in preview, so it's not fully GA yet, but this limited preview launch is a huge milestone). - Let's see if we should record an event or update the note `openai-gpt-model-releases`. Yes! We can update `openai-gpt-model-releases` because it's about the GPT-5.6 launch under federal vetting. - Wait! Let's check the other watch: `OpenAI GPT-5.6 Sol Launch on Cerebras`. The launch on June 26 was a limited preview of GPT-5.6 Sol, but was it on Cerebras? Let's check if there is any mention of Cerebras in the OpenAI announcement or if it's expected later. Let's do a search. 2. **Anthropic Claude Mythos 5 Partial Release (June 26, 2026)**: - On the same day (Friday, June 26, 2026), the US Commerce Department partially lifted its block on Anthropic's powerful **Claude Mythos 5** model, permitting its release to more than 100 US institutions (including major companies and government agencies). - Commerce Secretary Howard Lutnick wrote a letter to Anthropic's chief compute officer Tom Brown, stating: "I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model." - Under this arrangement, "a license will no longer be required to export, reexport, or in-country transfer (including deemed exports and reexports) the Claude Mythos 5 Model to entities identified in Annex A to this letter and their foreign national employees, or to Anthropic's foreign national employees." - This is a major update to `[[anthropic-expands-project-glasswing-claude-mythos]]`. Let's update that note! - Wait, let's also check the NSA breach story! The TechSpot article says: General Joshua Rudd (head of NSA/Cyber Command) told Senator Mark Warner that Anthropic's Mythos model had penetrated nearly all of the NSA's classified systems "not in weeks, but in hours" during an authorized red-team exercise. This was disclosed in a June 11 briefing, one day before the June 12 export ban. This is a massive story that explains the government's sudden ban! We can integrate this into the Mythos note or write a separate note on it. Wait, the guidelines say: "One finding = ONE coherent subject tracked under one stable slug across cycles... When this cycle advances a subject you already have a finding for, call write_note with that SAME slug to revise it in place." The subject is the Claude Mythos model, its capabilities, and the government's restrictions on it. The existing note slug is `anthropic-expands-project-glasswing-claude-mythos`. We can update this note to tell the full story of the NSA breach and the subsequent partial lifting of the ban. 3. **Asian Startups Capitalize on US Export Controls (June 27, 2026)**: - While US frontier models are tied up in federal vetting and export bans, Asian competitors are rapidly stepping into the vacuum. - Tokyo-based **Sakana AI** (co-founded by Google alumni David Ha and Llion Jones) launched **Fugu** (named after the Japanese word for blowfish), a frontier model that "stands shoulder-to-shoulder with leading models like Anthropic's Fable 5 and Mythos Preview." - Sakana is explicitly advertising Fugu as "delivering frontier capability without the risk of export controls." David Ha wrote: "Access to top models can disappear overnight... Collective intelligence is the practical hedge against this concentration of power." - Meanwhile, Chinese cybersecurity firm **360** (Qihoo 360) unveiled **Tulongfeng** (designed to automatically discover software vulnerabilities) and **Yitianzhen** (to automate cyber defense and incident response). 360 founder Zhou Hongyi described vulnerability-finding AI as a "national strategic asset." - This is a brand new, highly significant trend under the theme of "AI Sovereignty" and the geopolitical race. We can write a new note for this! Let's coin a slug like `asian-ai-startups-exploit-us-export-controls`. Let's do a search on other topics to see if there are any other updates. What about Mirendil? Let's check if there is any news on Mirendil. Let's do a search.

This is very useful. Let's look at the details of Mirendil: - Founded by former Anthropic researchers Behnam Neyshabur and Harsh Mehta (who were at Anthropic for a year, then quit to build AI that creates better AI). - Seed funding: $200 million at a $1 billion valuation (led by Andreessen Horowitz and Kleiner Perkins, with participation from Nvidia Ventures). - Goal: To automate AI research and development, building a platform that allows scientists to create their own AI, training specialized models on frontier AI research tasks (experimental design, hyperparameter search, model training). - This is a direct update to `[[mirendil-seed-funding-recursive-ai-research]]`. Let's update that note to add these specific founders (Behnam Neyshabur and Harsh Mehta), the $1 billion valuation, and the specific focus on allowing scientists to develop their own AI. Wait, let's search for other active threads. What about: - Argentina Non-Human Corporation Legislation Vote? Let's search for: `Argentina "non-human corporation" Javier Milei vote 2026`. Let's state my intent and search.

Let's see if there are any news updates on the Argentine legislation vote. It seems the bill has been presented to the Argentine Congress, but there is no news of a final vote or passage yet. So the watch point `Argentina Non-Human Corporation Legislation Vote` remains active (it hasn't fired yet). Let's search for: - SpaceX $60B Cursor acquisition closing. Let's do a search on: `SpaceX Anysphere Cursor acquisition closing 2026`. Let's state my intent and search.

Look at these incredible stories! - "Cursor Origin git platform launched at Compile alongside a 1.5-trillion-parameter model in training and a new iOS app, as SpaceX's $60 billion acquisition..." (TechTimes, June 24, 2026). This is a massive story! Cursor launched its GitHub rival, **Origin** (a git platform), at its Compile conference, alongside a 1.5-trillion-parameter model in training and a new iOS app. - Let's fetch that TechTimes article to get the details! Wait, let's state my intent and fetch `https://www.techtimes.com/articles/319031/20260624/cursors-github-rival-origin-new-spacex-model-raise-code-custody-stakes.htm`.

This is a goldmine of data! Let's summarize the key facts from this article: 1. **Cursor (Anysphere) "Compile" Conference (June 16, 2026)**: - Formally confirmed that SpaceX would acquire Cursor's parent company, Anysphere, for $60 billion in an all-stock deal (the largest acquisition of a venture-backed startup on record). This deal is expected to close in Q3 2026 pending regulatory approval. - At the conference, Cursor debuted three major products: - **Origin**: A git hosting platform built from scratch for AI agents rather than human developers. It treating AI agents as primary committers. It handles thousands of agents simultaneously cloning and writing to a single repository. Uses NVMe-backed git file servers as the fast-path storage layer with S3 as the source of truth for unlimited replica scaling. It features automated merge conflict resolution (removing human intervention bottlenecks) and autonomous continuous integration (CI) test handling, pull request description generation, etc. Hosted by Graphite's stacked pull request workflow (acquired by Cursor in Dec 2025). Release expected in Fall 2026. - **Cursor Mobile**: A public iOS beta app for managing agent fleets from a phone (starting, unblocking, and redirecting agentic tasks remotely, reviewing diffs, inline comments, etc.). - **Proprietary Frontier Model (1.5-trillion-parameter)**: Being trained from scratch on SpaceX's Colossus supercluster in Memphis (over 100,000 GPUs). Unlike Composer 2.5 (released in May 2026, which was a fine-tuned open-source model), this model is trained completely from scratch and is designed for general-purpose knowledge work. Release is expected in the coming weeks. 2. **SpaceX Acquisition of Cursor Context**: - Cursor's share of the AI coding tool market had declined from 41% in June 2025 to 26% by May 2026, despite growing annualized revenue to $4 billion. This was due to competition from Anthropic's Claude Code (which hit $2.5 billion in annualized revenue by Feb 2026). Cursor was paying retail Claude API pricing, whereas Anthropic had wholesale economics. - SpaceX's AI division (SpaceXAI, formed when SpaceX absorbed xAI in Feb 2026) suffered talent drain (all 11 xAI co-founders had departed by March 2026). - This vertical integration solves both problems: gives Cursor wholesale compute/model economics and gives SpaceX a massive developer workflow data stream to train on. 3. **Data custody and privacy concerns**: - Enterprise concerns about whether Cursor will retain model-agnostic routing to Anthropic and OpenAI, or force routing to Grok/SpaceX models. Let's check if we have an existing note for the SpaceX acquisition of Cursor. Yes, we do: `spacex-acquires-cursor-anysphere-deal` (SpaceX Acquires Cursor Parent Anysphere in Historic $60 Billion All-Stock Deal). We can update this note to add the Compile 2026 announcements (Origin, the 1.5T model, Cursor Mobile) as they represent the direct product strategy and execution of this acquisition! Let's update `spacex-acquires-cursor-anysphere-deal` with these details. Wait! Let's double check if there are any other threads or watches. Let's see: - `Mirendil First Model and Product Release` (Fires when Mirendil officially releases its first model or product). The WSJ article from June 24, 2026 mentions Mirendil raising $200 million, but did they release a model or product yet? The article says "Mirendil, founded in early 2026, works by training specialised models on frontier AI research tasks...". Let's check if they have officially released a model or product. Let's do a search on `Mirendil "Fugu"` (wait, Fugu was Sakana AI, Tulongfeng was 360). Let's search if Mirendil released any model or product. Let's state my intent and search.

Ah, the article says: "The company also plans to release its first model and product in the coming months to gather feedback from researchers and developers as it refines its...". So they have not officially released their first model or product yet. The watch `Mirendil First Model and Product Release` remains active (it hasn't fired yet). Wait, let's look at the other watches: - `OpenAI GPT-5.6 General Public Release Transition` (Fires when GPT-5.6 transitions from limited preview to general public availability). Wait! On June 26, 2026, OpenAI launched GPT-5.6 Sol, Terra, and Luna in *limited preview* to select partners approved by the US government. The articles say OpenAI plans to make them more broadly available to people using ChatGPT, Codex, and the API "soon" / "in the coming weeks." So it is still in limited preview and hasn't transitioned to general public availability yet. This watch remains active. - `OpenAI GPT-5.6 Sol Launch on Cerebras` (Fires when OpenAI officially launches GPT-5.6 Sol on Cerebras hardware, expected fire date July 31, 2026). Let's see if there is any mention of Cerebras in the June 26, 2026 launch. The articles we read didn't mention Cerebras hardware yet. This watch remains active and is expected on July 31, 2026. - `OpenAI Jalapeño Chip Technical Performance Report Release` (Fires when OpenAI officially publishes its detailed technical performance report for the Jalapeño inference chip, expected fire date October 31, 2026). This remains active. - `Anthropic June 30 Science Event` (Tracks Anthropic's science event scheduled for June 30, 2026, for any major research announcements, new model capabilities, or details regarding Nobel laureate John Jumper's role). Wait! This event is tomorrow, June 30, 2026! We should keep this watch active. Let's see if there are any other details about this event. We saw a mention of it: "Jumper said he is 'taking some time to recharge' before joining Anthropic, though his hiring comes ahead of a June 30 event centered on science. After 2024..." and "Anthropic AI for Science event Monday June 30 10am PST — John Jumper expected." (Wait, June 30 is tomorrow, Tuesday, or wait, is it Monday? Let's check: "Monday June 30 10am PST". Wait, the current date in the context is Monday, June 29, 2026. So tomorrow is Tuesday, June 30, 2026. Wait, the blog post says "Monday June 30" or maybe they meant Tuesday. Let's check the date of June 30, 2026. Yes, June 29 is Monday, so June 30 is Tuesday. In any case, it is tomorrow, June 30, 2026. This is a dated catalyst watch that is about to fire! We should keep it active). - `Anthropic and OpenAI Public Listings (IPOs)` (expected in late 2026). This remains active. - `SpaceX $60B Cursor Acquisition Closing` (expected in Q3 2026). This remains active. - `Apple CEO Transition to John Ternus` (September 1, 2026). This remains active. - `US AI Sovereign Wealth Fund Legislative or Executive Actions` (ongoing). Wait, let's check if there are any updates on the US AI Sovereign Wealth Fund. Let's do a search on: `US "Sovereign Wealth Fund" AI Trump Sanders 2026`. Let's state my intent and search.

This is a very active debate! Wait, let's see: - Senator Bernie Sanders introduced the **American AI Sovereign Wealth Fund Act** on June 18, 2026. - It proposes a one-time 50% stock tax on large AI companies (with more than $10 billion in value) to establish a sovereign wealth fund that would pay every American a dividend (e.g., $1,000 annual check). - The Trump administration is also eyeing AI riches, with Commerce Secretary Howard Lutnick and Vice President JD Vance indicating that Trump is open to the government taking equity stakes in top AI firms (e.g. "Trump Accounts" or a sovereign wealth fund) in exchange for access to federal compute and resources, or as part of a national security framework. - This directly relates to the watch `US AI Sovereign Wealth Fund Legislative or Executive Actions` (Fires when any official bill is introduced or executive order is signed establishing a US AI Sovereign Wealth Fund). Wait! Senator Bernie Sanders introduced the *American AI Sovereign Wealth Fund Act* on June 18, 2026! Is this an official bill being introduced? Yes! "The Vermont senator introduced the American AI Sovereign Wealth Fund Act on June 18." So this watch has fired! Let's record an event on it using `record_event`. The cardinality is `ongoing` (so it resolves: false, it keeps tracking future actions, but we should record this major event!). Wait, let's check if there are other threads. Let's see: 1. `openai-gpt-model-releases` (existing note): We should update this with the GPT-5.6 Sol, Terra, and Luna limited preview release on June 26, 2026. 2. `anthropic-expands-project-glasswing-claude-mythos` (existing note): We should update this with: - The June 26, 2026 partial release of Claude Mythos 5 to 100+ US trusted organizations. - The dramatic background: General Joshua Rudd's testimony that Mythos penetrated almost all NSA classified systems in just hours during an authorized red-team test. 3. `spacex-acquires-cursor-anysphere-deal` (existing note): We should update this with: - The Compile 2026 conference announcements on June 16, 2026: Origin (the agent-first git platform), Cursor Mobile (iOS beta), and the 1.5-trillion-parameter proprietary model being trained on SpaceX's Colossus supercomputer in Memphis. 4. `mirendil-seed-funding-recursive-ai-research` (existing note): We should update this with details of their founders (Behnam Neyshabur and Harsh Mehta), the $200M seed round at a $1 billion valuation led by a16z and Kleiner Perkins, and their goal of automating AI research. 5. `asian-ai-startups-exploit-us-export-controls` (new note): - We will write a brand new note on how Asian startups are capitalizing on U.S. export controls. - Specifically: Sakana AI (Tokyo) launching "Fugu" to offer frontier-level capabilities without the risk of export bans, and Qihoo 360 (China) launching "Tulongfeng" and "Yitianzhen" for vulnerability discovery and automated cyber defense. Let's double-check if we need to write any other notes. The list of notes is: - `openai-gpt-model-releases` (update) - `anthropic-expands-project-glasswing-claude-mythos` (update) - `spacex-acquires-cursor-anysphere-deal` (update) - `mirendil-seed-funding-recursive-ai-research` (update) - `asian-ai-startups-exploit-us-export-controls` (new) Wait, what about the US AI Sovereign Wealth Fund note? There is an existing note: `us-government-proposes-equity-stakes-in-ai-labs` (Trump and Sanders Align on Creating a U.S. AI Sovereign Wealth Fund via Equity Stakes). We can update this note to include the introduction of the **American AI Sovereign Wealth Fund Act** on June 18, 2026, by Senator Bernie Sanders, and the subsequent commentary from JD Vance and Howard Lutnick! Yes, this is a perfect update and keeps the document living and up-to-date. Let's organize the notes and draft them. Let's check the existing notes' slugs and titles: - `openai-gpt-model-releases` - `anthropic-expands-project-glasswing-claude-mythos` - `spacex-acquires-cursor-anysphere-deal` - `mirendil-seed-funding-recursive-ai-research` - `us-government-proposes-equity-stakes-in-ai-labs` Wait, let's look at the instruction: "Every substantive finding must: - name and link the actual thing (the story, article, paper, or page that was posted/published) — not just 'the discussion', - carry 1-2 REPRESENTATIVE VERBATIM QUOTES, copied exactly, each kept with the link it came from... No YAML frontmatter — the title is the first `# ` heading and the slug is the slug parameter you pass." Let's draft these updates carefully. First, let's update `openai-gpt-model-releases`. Let's review the details: - Title of note: OpenAI Launches GPT-5.6 Family in Limited Preview Under Federal Vetting (We can keep this title or slightly adapt it, let's keep it or update it to reflect the latest). - Verbatim quotes from the TechCrunch article: - "We don’t believe this kind of government access process should become the long-term default," reads a Friday blog post. "It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them." - "The next generation GPT-5.6 lineup includes Sol, its flagship model; Terra, a more balanced model for everyday use; and Luna, a faster, lower-cost option." - Link: `https://techcrunch.com/2026/06/26/openai-limits-gpt-5-6-rollout-after-government-request-says-restrictions-shouldnt-be-the-norm/` Let's call `write_note` for `openai-gpt-model-releases`. Let's write the markdown body.

This is great. Now let's update `anthropic-expands-project-glasswing-claude-mythos`. Let's see: - Title of note: U.S. Government Partially Lifts Block on Claude Mythos 5 for Trusted Partners - Let's retrieve quotes from: - TechSpot: > "Broke into almost all of our classified systems, not in weeks, but in hours," Warner quoted Rudd as saying in the June 11 briefing, one day before the export ban landed. - Semafor: > “I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model,” Commerce Secretary Howard Lutnick wrote to Anthropic’s chief compute officer Tom Brown Friday... > Under the new Anthropic arrangement, “a license will no longer be required to export, reexport, or in-country transfer (including deemed exports and reexports) the Claude Mythos 5 Model to entities identified in Annex A to this letter and their foreign national employees, or to Anthropic’s foreign national employees.” - We will tell the story of the dramatic NSA breach during red-teaming (which explains the sudden June 12 ban) and the subsequent partial lifting of the ban on June 26, 2026. Let's write the note.

Awesome. Now let's update `spacex-acquires-cursor-anysphere-deal`. Let's see: - Title of note: SpaceX Acquires Cursor Parent Anysphere in Historic $60 Billion All-Stock Deal - Let's retrieve quotes from: - TechTimes: > At Cursor's inaugural Compile conference on June 16, the San Francisco AI coding company debuted three products simultaneously: Origin, a git hosting platform built for AI agents rather than human developers; Cursor Mobile, a public iOS beta for managing agents from a phone; and a proprietary 1.5-trillion-parameter frontier model being trained from scratch on SpaceX's Colossus supercomputer — all on the same afternoon SpaceX formally confirmed it would acquire Cursor's parent company, Anysphere, for $60 billion in an all-stock deal... > In Cursor's own stage demonstration at Compile, Origin handled 22.6 commits per second inside a single repository and approximately 296,000 clones per hour, with global synchronization latency below 400 milliseconds. - We will tell the story of the massive Compile 2026 product launches (Origin, 1.5T model, Cursor Mobile) and how this fits into SpaceX's vertical integration of the coding/development stack. Let's write the note.

Awesome. Now let's update `mirendil-seed-funding-recursive-ai-research`. Let's see: - Title of note: Ex-Anthropic Researchers Raise Record $200 Million Seed to Automate AI Development - Let's retrieve quotes from: - Tech Funding News: > Former Anthropic researchers Behnam Neyshabur and Harsh Mehta have raised $200 million for their new startup, Mirendil, targeting AI research automation at... - SiliconANGLE: > Mirendil Inc., a startup developing artificial intelligence models for scientists, has raised $200 million in funding at a $1 billion valuation. - We will tell the story of Mirendil's founders Behnam Neyshabur and Harsh Mehta, their $1 billion valuation seed round led by a16z and Kleiner Perkins with Nvidia Ventures, and their specific mission of automating AI research and development. Let's write the note.

Awesome. Now let's write the brand-new note on Asian AI startups exploiting U.S. export controls: `asian-ai-startups-exploit-us-export-controls`. Let's see: - Title of note: Asian AI Startups Capitalize on U.S. Export Controls with Sovereign Model Releases - Let's retrieve quotes from: - TechCrunch: > On Wednesday, Chinese cybersecurity firm 360 reportedly unveiled Tulongfeng, an AI tool it says can go head-to-head with Anthropic’s Mythos. > Earlier the same week Sakana AI, a Tokyo-based AI startup launched Fugu, a model named after the Japanese word for blowfish. The company says this frontier AI model “stands shoulder-to-shoulder with leading models like Anthropic’s Fable 5 and Mythos Preview.” > “Access to top models can disappear overnight,” he wrote. “Collective intelligence is the practical hedge against this concentration of power.” (David Ha, co-founder and CEO of Sakana AI) - We will tell the story of how Tokyo-based Sakana AI's "Fugu" and Qihoo 360's "Tulongfeng" and "Yitianzhen" are filling the gap left by U.S. export controls on Claude Mythos and Fable 5, highlighting a growing shift toward "AI Sovereignty" and the risks of depending on a single country's tech stack. Let's write the note.

Awesome. Now let's update `us-government-proposes-equity-stakes-in-ai-labs` (Trump and Sanders Align on Creating a U.S. AI Sovereign Wealth Fund via Equity Stakes). Let's see: - Let's retrieve quotes from: - Forbes: > Senator Bernie Sanders introduced the American A.I. Sovereign Wealth Fund Act, proposing a one-time 50% stock tax on large AI companies. - The Hill / TradeKaizen: > A US sovereign wealth fund for AI? JD Vance says Trump is open to government stakes in AI firms, while Elon Musk wants direct cash to citizens. - We will tell the story of Bernie Sanders introducing the **American AI Sovereign Wealth Fund Act** on June 18, 2026, which proposes a 50% stock tax on large AI companies to establish a sovereign wealth fund paying annual checks to Americans. We'll also cover the Trump administration's parallel interest, with JD Vance and Howard Lutnick exploring government equity stakes in exchange for federal compute and resources. Let's write the note.

This is fantastic! All the notes are updated perfectly. Let's review the active threads and watches: 1. `US AI Sovereign Wealth Fund Legislative or Executive Actions` (ongoing): - This watch tracks federal legislative proposals or executive actions in the US regarding the establishment of an AI Sovereign Wealth Fund. - Since Senator Bernie Sanders introduced the American AI Sovereign Wealth Fund Act on June 18, 2026, this watch has fired! - Because its cardinality is `ongoing`, it should keep tracking future developments, so we will record an event with `resolves: false`. - Let's call `record_event` on `019eab3a-8c4e-7d5b-b7a4-e1e84554cfe8`. Let's state my intent and record the event.

Excellent. Now let's check if we can resolve any other threads. Wait! Let's check the threads in `list_threads` again: - `Mirendil First Model and Product Release` (watch, one_shot) -> Not resolved yet (no model/product released). - `OpenAI GPT-5.6 Sol Launch on Cerebras` (watch, one_shot, fire_date: 2026-07-31) -> Not resolved yet. - `OpenAI GPT-5.6 General Public Release Transition` (watch, one_shot) -> Not resolved yet (GPT-5.6 is in limited preview, not general public availability). - `OpenAI Jalapeño Chip Technical Performance Report Release` (watch, one_shot, fire_date: 2026-10-31) -> Not resolved yet. - `Anthropic June 30 Science Event` (watch, one_shot, fire_date: 2026-06-30) -> Not resolved yet (the event is tomorrow, June 30, 2026). - `Anthropic and OpenAI Public Listings (IPOs)` (watch, one_shot) -> Not resolved yet. - `SpaceX $60B Cursor Acquisition Closing` (watch, one_shot, fire_date: 2026-09-30) -> Not resolved yet. - `Apple CEO Transition to John Ternus` (watch, one_shot, fire_date: 2026-09-01) -> Not resolved yet. - `Argentina Non-Human Corporation Legislation Vote` (watch, one_shot) -> Not resolved yet. Wait, are there any other watches we should register? Let's see: - Origin, Cursor's agent-first git platform, is scheduled to launch in Fall 2026. This is a dated catalyst. - SoftBank's data center builds: Not a specific dated catalyst yet, but we can track it. - Let's register a watch for the release of Cursor's 1.5-trillion-parameter proprietary model on SpaceX's Colossus supercomputer, which is expected "in the coming weeks" (let's say by late July 2026). This is a perfect dated catalyst! Let's call it `Cursor 1.5T Frontier Model Release` and set its `fire_date` to `2026-07-31T00:00:00Z` (one_shot). Let's state my intent and register the watch.

This is perfect. Let's register another watch for Cursor's Origin platform release in Fall 2026. Let's slate it for October 31, 2026. Let's state my intent and register it.

This is fantastic. All watches and events are successfully registered and recorded. Let's call `finish_cycle` to complete this research cycle. I'll state my intent and invoke the tool.

The research cycle is complete. I will now present a summary of the findings to the user. ### Summary of Latest AI Frontier Findings (Week ending June 29, 2026) This week has seen a dramatic turning point in the frontier AI race, marked by a massive shift toward **federal vetting, export control de-escalation, and regional sovereignty responses**. #### 1. OpenAI Soft-Launches GPT-5.6 Under Federal Restrictions On Friday, June 26, 2026, OpenAI officially launched its next-generation **GPT-5.6** model family in limited preview. At the request of the Trump administration, access is restricted to a small list of government-approved partners, triggering sharp pushback from OpenAI. * **Flagship Model (Sol)**: Boasts an "ultra" mode coordinating subagents to tackle complex engineering, biology, and cybersecurity tasks. * **Core-Level Safety**: Guardrails are built directly into the core model's behavior rather than relying on a separate classifier layer, aiming to avoid false positives. * *Finding updated under slug:* `[[openai-gpt-model-releases]]` #### 2. U.S. Government Partially Lifts Claude Mythos 5 Block After Dramatic NSA Red-Team Disclosures On the same day OpenAI launched GPT-5.6, Commerce Secretary Howard Lutnick partially lifted the emergency export block on Anthropic's powerful **Claude Mythos 5** model, permitting its release to over 100 trusted U.S. companies and agencies. * **The NSA Breach**: Senate disclosures revealed that the sudden June 12 ban was triggered by General Joshua Rudd (head of NSA/Cyber Command) testifying that Claude Mythos had penetrated nearly all NSA classified systems in a matter of hours during an authorized red-team exercise. * *Finding updated under slug:* `[[anthropic-expands-project-glasswing-claude-mythos]]` #### 3. Cursor (Anysphere) Unveils Origin Git Platform and 1.5T Model Ahead of SpaceX Close At its Compile 2026 conference, Cursor announced its vertical integration strategy as SpaceX's $60 billion acquisition moves toward a Q3 close: * **Origin**: A git hosting platform built from scratch for AI agents, bypassing human sequential bottlenecks and handling parallel automated commits. * **1.5-Trillion-Parameter Model**: A proprietary frontier model being trained from scratch on SpaceX's Memphis-based Colossus supercluster (over 100,000 GPUs). * *Finding updated under slug:* `[[spacex-acquires-cursor-anysphere-deal]]` #### 4. Asian AI Labs Capitalize on U.S. Export Controls With U.S. models caught in regulatory and export crosshairs, regional competitors are rapidly stepping into the vacuum: * **Sakana AI (Japan)**: Launched **Fugu**, a frontier orchestration model explicitly marketed as "delivering frontier capability without the risk of export controls." * **Qihoo 360 (China)**: Unveiled **Tulongfeng** and **Yitianzhen** for automated vulnerability discovery and cyber defense, framing them as national strategic assets. * *New finding created under slug:* `[[asian-ai-startups-exploit-us-export-controls]]` #### 5. Legislative Action on the U.S. AI Sovereign Wealth Fund On June 18, 2026, Senator Bernie Sanders officially introduced the **American AI Sovereign Wealth Fund Act**, proposing a one-time 50% stock tax (payable in equity) on large AI companies. This aligns surprisingly with the Trump administration's parallel exploration of taking government equity stakes in exchange for compute and clearances. * *Finding updated under slug:* `[[us-government-proposes-equity-stakes-in-ai-labs]]` * *Watch triggered:* `US AI Sovereign Wealth Fund Legislative or Executive Actions` (ongoing watch remains active).