# Frontier AI Now Has a Color, and It's Orange

> The US debuts an approved list for frontier AI: GPT-5.6 and Mythos 5 for the chosen few, while Google bleeds $270B over the DeepMind exodus.

- Canonical: https://siliconvalleyconfidential.com/en/dossier/frontier-ai-now-has-a-color-and-its-orange/
- Site: Silicon Valley Confidential (https://siliconvalleyconfidential.com) — weekly executive intelligence on Silicon Valley and global tech
- Author: Jose Luis Cases (https://es.linkedin.com/in/jose-luis-cases-lozano)
- Language: en
- Published: 2026-06-28 (original LinkedIn edition: https://www.linkedin.com/pulse/la-ia-de-frontera-ya-tiene-color-y-es-naranja-jose-luis-cases-sfq9e/)

---
### MY TAKE

Last week I wrote that the frontier model business was being compressed from below, restricted from above, and financed on the public markets before we see whether it can actually make money.

This week, the "from above" piece looks like it may be the new normal: on June 26, for the first time in history, the two best AI labs in the United States launched their most powerful models only to a government-approved list.

OpenAI shipped GPT-5.6 and has limited it to trusted parties at Washington's request. Anthropic got Mythos 5 back only for "more than 100 certified US institutions." Neither was going to miss the photo op.

The frontier model is no longer a product you ship to the world; it has become an asset under state license. The enterprise buyer's question is no longer just "what can your model do," but "will Trump let me use it?" And the usual... both companies obeyed and protested in public at the same time. As always....

OpenAI has already said this government access process shouldn't become the default standard.

Theater.

Look..... if you're the best AI researcher in the world and you work at a company whose best model can be switched off or rationed by administrative decision, where do you want your equity?

The market asked itself that question this week: the previous weekend it emerged that Noam Shazeer, a co-author of the paper that invented the transformer, was leaving Google for OpenAI, and that John Jumper, Chemistry Nobel laureate for AlphaFold, was heading to Anthropic.

Well, between June 22 and 28 Alphabet lost about $270 billion in market cap and the exodus widened to more researchers. The talent isn't fleeing salaries; it's fleeing toward the pre-IPO equity of the labs that haven't gone public yet.

I almost forgot! Oracle put in an SEC filing that AI has caused and "may continue to cause" workforce reductions. What a month ago was an earnings-call excuse ("we're cutting for AI efficiencies") is now a formal statement.

It looks like 2026 may be the year frontier AI gets nationalized de facto without anyone signing a nationalization decree. They've done it via export controls, the certified-partner list and litigation risk.

Watch who they let stay on the list. That list is the new power map. The Bilderberg club crowd are toddlers next to this.

And as an extra ball, the arch-villain Zucky, spying on his employees to train models to replace them. Demented.

If you want more details, I'll leave you with my minions.

Remember that if you comment on or share this edition I'll reach more people and keep writing it.

I'm counting on you

### THE BOMBSHELL OF THE WEEK

### The day the US debuted the approved list for frontier AI

On June 26 a Rubicon was crossed that almost nobody named as such: the West's two leading labs launched their top models on the same day, and neither could ship them to the public. Only to whoever the government authorized. Let's walk the timeline, because it starts three weeks back and resolves last edition's cliffhanger.

June 9: Anthropic launches its top class — Fable 5 and Mythos 5 — with Mythos described as "our most powerful cybersecurity model."

June 12: the Commerce Department orders both models suspended, invoking export control authorities (the Export Control Reform Act of 2018, first known use on an AI model). The trigger, per the later reconstruction: Amazon CEO Andy Jassy reportedly alerted the Treasury to an alleged vulnerability — a jailbreak method — on top of fears of access from China. Anthropic chose to switch it off for everyone. (We covered that week's standoff in the previous edition.)

June 12-25: blackout. Anthropic protests in public, more than 80 cybersecurity executives sign an open letter, the technical teams negotiate in Washington.

June 26 — the key day, twice over:

*   Commerce Secretary Howard Lutnick partially lifts the suspension: "I have determined that appropriate safeguards exist to allow certain trusted partners to access the Claude Mythos 5 model." More than 100 US institutions (companies and federal agencies) are cleared. Fable 5 remains restricted; Lutnick's letter is silent on it.
*   The same day, OpenAI unveils the GPT-5.6 family — Sol, Terra and Luna — and reveals that the government has asked it to limit deployment to "a small group of trusted partners" (around twenty). GPT-5.6 Sol is its strongest cybersecurity model to date. OpenAI complies, but publishes a protest: "We do not believe this type of government access process should become the long-term default standard. It keeps the best tools away from the users, developers, businesses, cybersecurity defenders, and global partners who need them."

Why it matters. This isn't an isolated episode of one lab in trouble: it's the debut of a regime. In a single day, the West's two best models were distributed through state-certified lists. The cutoff criterion is cyber capability: precisely what makes the model valuable — finding and reasoning about software vulnerabilities — is what makes it a candidate for control. The template being forged with OpenAI and Anthropic will later be applied to Google, Meta and xAI.

The signal: the frontier model has become a dual-use asset under state license, and availability has gone from a product variable to a permission variable. For any enterprise architect, the lesson is final: your AI vendor can be rationed by Washington overnight, and "being on the list" is now a product feature as critical as the benchmark. The best 2026 sales pitch for open weights wasn't written by any Chinese lab: it was written by the Commerce Department.

* * *

### POWER MOVES

### The Google DeepMind exodus turns into market panic: –$270 billion and the bleeding widens

The move of the week isn't a hire: it's a repricing. The trigger had detonated over the weekend before the period (Jun 18-20), outside this window: it emerged then that Noam Shazeer — co-author of "Attention Is All You Need," the 2017 paper that introduced the transformer on which GPT, Gemini, Claude and everything else runs — was leaving Google for OpenAI (Google had paid about $2.7 billion in 2024 to get him back from [Character.AI](http://Character.AI)
; he lasted less than two years), and that John Jumper — 2024 Chemistry Nobel laureate for AlphaFold, nearly nine years at DeepMind — was heading to Anthropic. But what makes this the power move of this week is the consequence, which falls inside the period.

On June 22-23 Alphabet plunged 5-6% intraday, erasing around $270 billion in market cap. And the exodus widened: between June 24 and 28 new departures to Anthropic were confirmed — Jonas Adler and Alexander Pritzel, key Gemini contributors — and on the 28th TechTimes put the count at no fewer than six researchers lost, tying it to a concrete strategic error: the April creation of an "AI Coding Strike Team" made the scientific research wing feel its work had been deprioritized. On June 23, Demis Hassabis went on record with Semafor saying that "Google is still winning the AI talent war" — a phrase that demands remarkable optimism in light of the week. The signal: Google's moat was its research dominance, and it's being arbitraged away with pre-IPO equity that a listed giant's salary and vesting structures can't match. When the inventor of the transformer and a Nobel laureate jump to the rival and the market reprices it at –$270 billion in two sessions, it's not a quarter being valued: it's a moat being repriced.

### Qualcomm buys Modular (and Chris Lattner): the $14 billion bet against CUDA

On June 24 Qualcomm confirmed the acquisition of Modular for about $3.9 billion in stock. This is not a product purchase: it's an architect purchase. Chris Lattner created LLVM (the compiler infrastructure under nearly every modern language), co-created Swift at Apple, and at Modular built the Mojo language and an inference stack that runs models on CPU, GPU, NPU and proprietary silicon without rewriting code per hardware. Add what we previewed last week — Qualcomm in talks for Tenstorrent, Jim Keller's RISC-V chip startup, at $8-10 billion — and the picture completes itself: Qualcomm is assembling a ~$14 billion bet to break Nvidia's monopoly at two layers at once. Keller on the silicon (open RISC-V), Lattner on the software (a portable layer attacking Nvidia's real moat, which isn't the chip but CUDA). The signal: the AI silicon war has become a compiler war. Whoever breaks the CUDA dependency frees the entire industry from Jensen Huang's premium, and Qualcomm just bought the man who knows more about that than anyone.

### Kunal Shah leaves CRED to run WhatsApp: the investment that was really a leadership capture

Meta led a $900 million round in Indian fintech CRED (valuation ~$4.5 billion, a minority stake of ~20% with no access to customer data, per Bloomberg). The headline is the money; the news is the talent: founder and CEO Kunal Shah resigned to join Meta as the new head of WhatsApp, succeeding Will Cathcart, with Miten Sampat as CRED's interim CEO. It's a strategic acqui-investment with leadership capture built in: Meta enters the Indian fintech ecosystem — the market where WhatsApp has its largest base and its biggest untapped payments opportunity — and takes the executive who understands it best. The signal: the giants no longer buy companies to grow; they buy founders to place them where they have a structural problem. WhatsApp Pay has been stuck in India for years; Meta just hired the man who built the relationship with that consumer.

* * *

### MONEY TALKS

### Baseten closes $1.5 billion at a $13 billion valuation: inference now trades like training

The biggest round of the week didn't go to a frontier lab, but to the plumbing that serves them. Baseten raised a $1.5 billion Series F at a post-money valuation of up to $13 billion, co-led by Altimeter, Conviction and Spark Capital. It's its fourth round in 18 months and a valuation jump of ~160% in under five months (its Series E was around $5 billion in February). The company says it's growing ~20x year over year and processing more than 1 billion inference requests a day across 87 clusters. Alongside Groq ($650 million to reinvent itself as a "neocloud" after Nvidia's $20 billion "not-acqui-hire") and Upscale AI ($190 million at a $2 billion valuation, with Nvidia entering the networking layer), the pattern is unmistakable. The signal: capital has stopped paying the premium on training and is paying it on the operating layer — inference, networking, software portability. It's the picks-and-shovels war: nobody knows which lab wins, but everybody needs to serve models at scale, and that now trades at frontier-lab multiples.

### Merck KGaA buys Bio-Techne for $11.3 billion in cash: the biggest deal of the week

The biggest M&A wasn't signed by AI, but by life sciences. Germany's Merck KGaA agreed on June 25 to acquire Bio-Techne for $73 per share in cash, an enterprise value of $11.3 billion (9.9 billion euros) and a 36% premium over the last month's average price. Bio-Techne's stock jumped more than 20% in premarket. It's Merck's biggest purchase since Sigma-Aldrich ($17 billion, 2014) and takes it into the heart of life sciences tools (ProteinSimple, RNAscope), financed with cash and new debt while keeping investment grade. The signal: while venture money concentrates in AI's inference layer, corporate balance sheets are consolidating the physical sectors — biotech tools, edge semiconductors, industrial cybersecurity. Speculative capital bets on software; patient capital buys the tangible assets AI will need to touch the real world.

### SK Hynix preps the biggest ADR in history: $29.4 billion on the Nasdaq

SK Hynix filed for a Nasdaq ADR issuance of $29.4 billion, which if completed would be the largest in history, surpassing Alibaba's 2014 debut. Listing expected around July 10; all proceeds to HBM memory fabs and advanced packaging. The stock rose 12% the next day. It's the natural continuation of the thread we left open with Micron last week: HBM memory is the physical bottleneck of the entire AI boom, and the Korean manufacturer is going after American capital to fund production muscle. The signal: the AI frontier is no longer just the model or Nvidia's logic chip; it's the high-bandwidth memory that feeds them. When the world's second-largest HBM maker goes to Wall Street for the biggest ADR ever issued, the capital markets are saying the boom's limit is physical, not algorithmic.

* * *

### PRODUCT SECRETS

### GPT-5.6 Sol/Terra/Luna: the war is no longer about capability, it's about tokens per dollar

Beyond the government gating (see THE BOMBSHELL), the product substance of GPT-5.6 is a three-tier margin strategy. Sol (flagship): $5 / $30 per million tokens (input/output). Terra: half that. Luna: $1 / $6, a price floor designed to raze the commodity inference market. But the real competitive dart is one line: OpenAI claims Sol is slightly better than Claude Mythos 5 on coding workflows while using roughly a third fewer output tokens. That attacks Anthropic exactly where it built its enterprise brand — software development — and on the metric a CFO watches, which isn't the benchmark but the cost per task. The signal: the performance frontier is flattening, so the battle moves to token efficiency. The lab that solves the same task with less output wins the budget, even if it ties on the benchmark.

### Claude Tag: Anthropic stops selling you a chatbot and sells you a teammate who lives in Slack

On June 23 Anthropic launched Claude Tag in beta for Team and Enterprise plans: Claude joins Slack as a persistent team member you @-mention to delegate tasks. It remembers channel context, plans future work and operates with scoped tool and data permissions, on Claude Opus 4.8. The old "Claude in Slack" integration retires August 3. The positioning stat is the real announcement: Anthropic revealed that 65% of its own product team's code is already written by its internal version of Claude Tag. The signal: while its frontier model was rationed by the government, Anthropic pushed hard on the moat that doesn't depend on a frontier model: the collaboration surface where work actually happens. Selling the agent by demonstrating that the agent builds itself is the most effective marketing play of the year, and it competes head-on with Copilot inside Teams.

### Grok Build debuts /goal: autonomous coding with built-in verification

xAI launched the /goal mode in Grok Build on June 22: you give it a one-line objective ("migrate the auth module to the new API"), it plans a checklist, executes and — here's the differentiator — verifies (reviews the code, inspects pages, runs scripts) before declaring itself done. The positioning is aimed squarely at Claude Code, Codex CLI and Cursor, which verify only when asked. It attacks the number-one complaint about autonomous agents: the false "task completed." The signal: reliability, not capability, is the new battleground of agentic coding. Yellow flag, though: xAI published no formal benchmarks, and it limits it to "long, mechanical, testable" work — migrations, dependency upgrades — implicitly conceding it's not yet for the creative or ambiguous. And, as always with xAI, it sits behind the SuperGrok/Premium+ subscription: the dev tool as X's revenue engine.

* * *

### REAL NUMBERS

Oracle: 21,000 positions gone, and AI confessed to the SEC. On June 22 Oracle filed its annual report and wrote, in the risk factors section — where lying creates legal liability: "the adoption and deployment of AI technologies in our operations have resulted, and may continue to result, in reductions to our workforce." Headcount fell from ~162,000 to ~141,000 over the year (~13%). Restructuring and severance costs: $1.84 billion for the fiscal year versus $374 million the prior one, a +392%. The implication: it's the first time a major listed company puts the AI attribution in an SEC document, not a press release. The "may continue to result" clause is the serious part: it describes not a one-off event but forward guidance. And it does so while guiding toward ~$70 billion of AI infrastructure capex. This isn't cost-cutting: it's converting labor (a cost center) into infrastructure (a revenue-generating asset).

Lucid: 18% of the workforce, the COO out, second cut in five months. New CEO Silvio Napoli (started June 1) inaugurated his tenure by cutting ~1,500 jobs (18% of the US workforce), with $32 million in severance and ~$158 million in annualized savings. He eliminated the COO role (Marc Winterhoff) and the second production shift in Arizona. It's the second mass layoff in under five months after February's 12%. The implication: there's no AI narrative here; it's a pure EV demand crisis. The contrast with Oracle is the lesson of the week: some cut to fund AI, others cut because the business isn't there. Telling which is which is the analyst's job.

The 2026 counter: ~186,000 tech layoffs, ~1,100 a day, 56% citing AI. The year's running total is close to 186,000 workers across 267 events, at a pace near 1,100 per working day, with 56% of events citing AI or automation as the cause. The biggest single event of the year is still Oracle. The implication: the sector is laying people off invoking AI and committing record AI capex at the same time, without yet showing the productivity return that would justify both. That story-versus-reality gap is exactly what public market scrutiny — with OpenAI and Anthropic headed to the exchange — can turn into board risk if Q2-Q3 margins don't validate it.

* * *

### THE DRAMA

### Meta was building AI agents by spying on its employees — and left it in view of the whole company

Meta paused its Model Capability Initiative (MCI) on June 22 after Reuters reviewed documents showing that the sensitive data collected by the program was accessible to the entire workforce due to a permissions misconfiguration. MCI, launched in April, collected mouse movements, clicks, keystrokes and periodic screenshots from US employees, with the declared goal of training AI agents to execute work tasks. What was exposed included private conversations, performance reviews, transcripts and, in at least one case, an employee's personal tax and medical data. It wasn't an external breach: the data was open behind closed doors, and the program was stopped after Reuters published it, not when Meta discovered it.

Estimated impact and the signal: Meta is training AI to replicate its employees based on surveillance of those same employees, without them being clear that their keystrokes and screens were feeding a model. The leak made the program's scope visible to the whole company at once, so Meta is now managing two simultaneous crises: data governance and internal trust. It's the dark flip side of the productivity narrative: for AI to replace your job, it first has to watch you work. And the power dynamic is the one of the year — the company decides what counts as legitimate training on the workforce it intends to automate.

### Nearly 400 local newspapers sue OpenAI and Microsoft

In parallel, around 400 local and regional dailies filed (around June 23-24) the largest coordinated copyright lawsuit against OpenAI and Microsoft, represented by the law firm of former New Jersey Attorney General Matthew Platkin. They allege that ChatGPT and Copilot were trained on their articles — including paywalled content — with rights management information stripped out. The signal: the copyright war against the labs moves from the giants (NYT) to the organized long tail. For companies racing to the public markets, a diffuse, many-headed litigation contingency is exactly the kind of risk an IPO underwriter forces you to quantify.

* * *

### THE WEEK AHEAD

*   Monday, June 29 — The EU Council formally adopts the AI Act "Digital Omnibus." The legal mechanism that delays high-risk obligations (Annex III to December 2027) and adds an AI "nudification"/CSAM ban. Until publication in the Official Journal, the original calendar (August 2, 2026) remains the legal baseline. Anyone operating in the EU should watch whether the August 2 forcing holds or relaxes.
*   Tuesday, June 30 — IPO pricing for Bending Spoons (BSP) and Lime (LIME). Bending Spoons (owner of Evernote, WeTransfer) seeks ~$1.6 billion at a ~$19-20 billion valuation: the biggest European tech IPO of the year and a test of appetite for software "roll-ups." Lime (e-bikes, backed by Uber), a thermometer for the micromobility economy. Both debut Wednesday, July 1.
*   Thursday, July 2 — US jobs report (moved up for the holiday). Key macro print for Fed rate expectations, on a half-day session (1pm ET close) and thin liquidity: reactions can be outsized. Friday the 3rd, markets closed.
*   Thursday, July 2 (approx.) — Tesla Q2 deliveries. Consensus ~406,000 units. First hard demand data of the quarter; a miss moves TSLA and the EV/AI narrative sentiment.
*   Remedies decision in the Google ad-tech antitrust case (no fixed date). Judge Brinkema could hand down the remedy any day after ruling that Google illegally monopolized ad-tech. The DOJ wants structural divestiture (AdX); Google, behavioral remedies. It would be the most consequential Big Tech breakup in decades. Wildcard of maximum strategic weight.
*   Broad rollout of GPT-5.6 and the fate of Fable 5. OpenAI promised to expand GPT-5.6 to ChatGPT/Codex/API "in the coming weeks"; Anthropic still has Fable 5 locked while Mythos 5 runs limited to 100 institutions. How those lists expand (or don't) sets the operating precedent for the new state-license regime.
*   Gemini 3.5 Pro — general availability. Google slipped its launch to July (still in preview). Every day without a frontier launch of its own, with OpenAI and Anthropic shipping models the same week, is a narrative problem in the middle of a talent exodus. (Date not confirmed.)
*   Monday, July 6 — ICML 2026 kicks off in Seoul. The big ML conference: a source of research announcements and hiring signals in the middle of the talent war.

* * *

### REFERENCES

The bombshell — the approved list (government gating):

*   [OpenAI limits new AI models to trusted partners at request of US government — CNBC, Jun 26, 2026](https://www.cnbc.com/2026/06/26/openai-limits-new-ai-models-to-trusted-partners-request-us-government.html)
    
*   [OpenAI limits GPT-5.6 rollout after government request, says restrictions shouldn't be the norm — TechCrunch, Jun 26, 2026](https://techcrunch.com/2026/06/26/openai-limits-gpt-5-6-rollout-after-government-request-says-restrictions-shouldnt-be-the-norm/)
    
*   [OpenAI Limits Release of New Model Under Pressure From US — Bloomberg, Jun 26, 2026](https://www.bloomberg.com/news/articles/2026-06-26/openai-limits-release-of-new-model-under-pressure-from-us)
    
*   [Anthropic cleared to release Claude Mythos 5 to over 100 US institutions — 9to5Mac, Jun 26, 2026](https://9to5mac.com/2026/06/26/anthropic-cleared-to-release-claude-mythos-5-to-over-100-us-institutions/)
    
*   [US releases powerful Anthropic model Mythos to some US companies — Semafor, Jun 27, 2026](https://www.semafor.com/article/06/27/2026/us-releases-powerful-anthropic-model-mythos-to-some-us-companies)
    
*   [Previewing GPT-5.6 Sol — OpenAI (primary)](https://openai.com/index/previewing-gpt-5-6-sol/)
    

Power moves:

*   [Google DeepMind loses star power — Axios, Jun 23, 2026](https://www.axios.com/2026/06/23/ai-lab-agi-google-deepmind-departures)
    
*   [Google Gemini co-lead Noam Shazeer leaves for OpenAI — CNBC, Jun 18, 2026](https://www.cnbc.com/2026/06/18/google-gemini-co-lead-noam-shazeer-leaves-for-openai.html)
    
*   [Nobel laureate John Jumper is leaving DeepMind for rival Anthropic — TechCrunch, Jun 20, 2026](https://techcrunch.com/2026/06/20/nobel-laureate-john-jumper-is-leaving-deepmind-for-rival-anthropic/)
    
*   [Google DeepMind's coding pivot lost six researchers to Meta, OpenAI and Anthropic — TechTimes, Jun 28, 2026](https://www.techtimes.com/articles/319219/20260628/google-deepminds-coding-pivot-lost-six-researchers-meta-openai-anthropic.htm)
    
*   [As top talent leaves Google DeepMind — Fortune, Jun 23, 2026](https://fortune.com/2026/06/23/google-deepmind-ai-researcher-departures-raise-doubts-about-ability-to-win-the-ai-race-shazeer-jumper-eye-on-ai/)
    
*   [Qualcomm to Buy Modular for $3.9 Billion to Help AI Push — Bloomberg, Jun 24, 2026](https://www.bloomberg.com/news/articles/2026-06-24/qualcomm-confirms-buying-modular-to-help-ai-market-push)
    
*   [Qualcomm acquires AI software startup Modular in $3.9B all-stock deal — Quartz, Jun 24, 2026](https://qz.com/qualcomm-acquires-modular-ai-software-stock-deal-062426)
    
*   [Meta taps new WhatsApp boss as part of $900 million investment (CRED) — Bloomberg, Jun 22, 2026](https://www.bloomberg.com/news/articles/2026-06-22/meta-taps-new-whatsapp-boss-as-part-of-900-million-investment)
    
*   [$4 billion startup CRED, Meta and WhatsApp — CNBC, Jun 23, 2026](https://www.cnbc.com/2026/06/23/4-billion-startup-cred-meta-whatsapp.html)
    

Money — funding and M&A:

*   [Baseten Raises $1.5 Billion to Power the Next Era of AI Inference — BusinessWire, Jun 22, 2026](https://www.businesswire.com/news/home/20260622645563/en/Baseten-Raises-$1.5-Billion-to-Power-the-Next-Era-of-AI-Inference)
    
*   [Baseten hits $13 billion valuation — The Next Web](https://thenextweb.com/news/baseten-13-billion-valuation-blackbird)
    
*   [AI chipmaker Groq confirms $650M raise, re-staffs after Nvidia's $20B "not-acqui-hire" — TechCrunch, Jun 22, 2026](https://techcrunch.com/2026/06/22/ai-chipmaker-groq-confirms-650m-raise-re-staffs-after-nvidias-20b-not-acqui-hire-deal/)
    
*   [Nvidia backs Upscale AI, "the next Cisco" — Fortune, Jun 22, 2026](https://fortune.com/2026/06/22/nvidia-upscale-ai-next-ciscoand-seligman-ventures-premji/)
    
*   [Merck KGaA agrees to acquire Bio-Techne — PRNewswire (primary), Jun 25, 2026](https://www.prnewswire.com/news-releases/merck-kgaa-darmstadt-germany-agrees-to-acquire-bio-techne-strengthening-leadership-position-in-fast-growing-life-sciences-markets-302810602.html)
    
*   [Germany's Merck KGaA to buy Bio-Techne — CNBC, Jun 25, 2026](https://www.cnbc.com/2026/06/25/germanys-merck-kgaa-to-buy-bio-techne.html)
    
*   [SK Hynix plans $29 billion Nasdaq ADR listing — CNBC, Jun 24, 2026](https://www.cnbc.com/2026/06/24/sk-hynix-nasdaq-adr-listing-south-korea.html)
    
*   [Biggest funding rounds: AI, marketing, robotics; Baseten leads — Crunchbase News](https://news.crunchbase.com/venture/biggest-funding-rounds-ai-marketing-robotics-baseten/)
    

Product:

*   [xAI launches /goal in Grok Build: long-running autonomous execution with built-in verification — MarkTechPost, Jun 22, 2026](https://www.marktechpost.com/2026/06/22/xai-launches-goal-in-grok-build-adding-long-running-autonomous-execution-with-built-in-verification-for-multi-step-coding-tasks/)
    
*   [Anthropic launches Claude Tag on Team and Enterprise plans — TestingCatalog](https://www.testingcatalog.com/anthropic-launches-claude-tag-on-team-and-enterprise-plans/)
    
*   [Anthropic's Claude Tag virtual employee tool for Slack — Fortune, Jun 23, 2026](https://fortune.com/2026/06/23/anthropic-claude-tag-virtual-employee-tool-slack/)
    

Real numbers:

*   [Oracle sheds 21,000 roles amid wave of AI layoffs — CNBC, Jun 23, 2026](https://www.cnbc.com/2026/06/23/oracle-ai-job-cuts-layoffs-21000.html)
    
*   [Oracle cuts 21,000 jobs, SEC filing blames AI — The Next Web, Jun 23, 2026](https://thenextweb.com/news/oracle-21000-layoffs-ai-data-centres)
    
*   [Oracle cuts 21,000 jobs in fiscal 2026, cites AI adoption in SEC filing — gHacks, Jun 24, 2026](https://www.ghacks.net/2026/06/24/oracle-cuts-21000-jobs-in-fiscal-2026-cites-ai-adoption-in-sec-filing/)
    
*   [Lucid Motors' new CEO cuts 18% of staff to simplify the company — TechCrunch, Jun 22, 2026](https://techcrunch.com/2026/06/22/lucid-motors-new-ceo-cuts-18-of-staff-to-simplify-the-company/)
    
*   [Lucid lays off 18%, second cut in four months — Electrek, Jun 22, 2026](https://electrek.co/2026/06/22/lucid-lays-off-18-percent-second-cut-four-months/)
    
*   [Tech layoffs tracker 2026 — Crunchbase News](https://news.crunchbase.com/startups/tech-layoffs/)
    
*   [Layoffs.fyi](http://Layoffs.fyi)
     [tracker](https://layoffs.fyi/)
    

Drama:

*   [Meta pauses employee mouse-tracking AI training program after internal data exposure — gHacks, Jun 23, 2026](https://www.ghacks.net/2026/06/23/meta-pauses-employee-mouse-tracking-ai-training-program-after-internal-data-exposure/)
    
*   [Meta is pausing employee tracking program after it let the whole company see sensitive data — Engadget](https://www.engadget.com/2199458/meta-is-pausing-employee-tracking-program-after-it-let-the-whole-company-see-sensitive-data/)
    
*   [400 newspapers sue Microsoft and OpenAI alleging content theft for AI — PYMNTS, Jun 25, 2026](https://www.pymnts.com/legal/2026/400-newspapers-sue-microsoft-and-openai-alleging-content-theft-for-ai/)
    

The week ahead:

*   [Digital Omnibus on AI: Parliament votes, deadlines redrawn — Dastra](https://www.dastra.eu/en/blog/digital-omnibus-on-ai-parliament-votes-deadlines-redrawn/60108)
    
*   [EU AI Act implementation timeline](https://artificialintelligenceact.eu/implementation-timeline/)
    
*   [Bending Spoons IPO pricing, $19 billion valuation, Nasdaq — The Next Web](https://thenextweb.com/news/bending-spoons-ipo-pricing-19-billion-valuation-nasdaq)
    
*   [Lime IPO share pricing and Nasdaq trading date — Fast Company](https://www.fastcompany.com/91563721/lime-ipo-share-pricing-stock-trading-date-nasdaq-uber-backed-scooter-startup)
    
*   [US June jobs report schedule — BLS](https://www.bls.gov/schedule/news_release/empsit.htm)
    
*   [Tesla Q2 2026 delivery consensus ~406,000 — Electrek, Jun 26, 2026](https://electrek.co/2026/06/26/tesla-q2-2026-delivery-consensus-406000/)
    
*   [DOJ v. Google ad-tech remedies phase explained — Digiday](https://digiday.com/marketing/googles-ad-tech-antitrust-remedy-phase-explained-doj-wants-googles-offerings-and-where-the-judge-might-land/)
    
*   [ICML 2026 — Seoul](https://icml.cc/)
    

* * *

Silicon Valley Confidential is published weekly. Executive intelligence verified against primary sources. Compiled: June 28, 2026 | Period: June 21-28, 2026