Anthropic Struggles to Resolve Mythos…

1. Key Themes

AI Governance & Export Controls Are Now an Operational Risk for AI Companies

Anthropic was given a 90-minute ultimatum to comply with Trump administration export controls on its Fable and Mythos models, forcing the company to suspend both systems for all users and dispatch senior technical staff to Washington. This signals that government intervention can create instantaneous, enterprise-wide product outages — a material risk for AI companies and their customers alike.

"Anthropic was given 90 minutes to comply with Trump administration export controls on its Fable and Mythos models, prompting the company to suspend both systems for all users and fly senior technical staff to Washington as it tries to resolve a clash that has rattled the AI industry."

AI Agent Security & Identity Management Is a High-Conviction Investment Theme

Multiple large funding rounds this week clustered around the same problem: controlling what AI agents can access, modify, and authenticate. Arcade.dev ($60M Series A), NewCore ($66M), and Apono (acquired by 1Password for $250–300M) all address authorization, identity, and access management for AI agents across enterprise systems.

Arcade.dev "develops software for controlling which enterprise apps, databases, and tools AI agents are authorized to access or modify." NewCore "manages identities, permissions, and access lifecycles for human employees and AI agents across enterprise systems." Apono's "AI-powered platform manages just-in-time access permissions across cloud infrastructure, enterprise applications, and corporate databases."

The AI Wealth Gap Is Creating Societal Powder Keg Conditions

As mass layoffs accelerate — with AI cited as the cause for three consecutive months — a narrow group of insiders is minting billionaires overnight. The article frames this as "combustible," with structural inequality deepening in real time.

"What makes this combustible is that at the very moment that tens of thousands of workers are being shown the door, a small cohort of AI insiders is becoming wealthy on a scale that's hard to comprehend." "Tech layoffs hit their highest single month in two years last month, with nearly 40,000 cuts, and AI was the most-cited reason for layoffs across every industry for the third month running."

Streaming Consolidation Is Accelerating Around Ad-Supported Scale

Fox's $25B acquisition of Roku signals that distribution reach and ad-supported inventory — not just content — are the decisive battleground in the streaming wars.

"Fox agreed to buy Roku in a roughly $25 billion deal, combining its Tubi, Fox One, and Fox Nation streaming businesses with the largest connected-TV streaming platform as media companies race for scale in ad-supported streaming."

Defense & Dual-Use Tech Is Attracting Durable Capital

Maritime Robotics (autonomous sea drones for energy, defense, and environmental monitoring) closed a $32.5M round despite being 21 years old, suggesting patient capital is now flowing into defense-adjacent hardware that previously struggled to attract VC interest.

Maritime Robotics "develops autonomous surface vessels, navigation systems, and modular sea drone platforms for offshore energy, surveying, environmental monitoring, fisheries, and defense operations."

2. Contrarian Perspectives

AI is largely a cover story for layoffs rooted in pandemic-era overhiring, not genuine automation displacement.

Marc Andreessen — himself a major AI booster — argues that most large companies were chronically overstaffed long before AI arrived, and are now using it as politically convenient cover. The Block/Jack Dorsey episode is the clearest illustration: Dorsey initially credited AI for enabling "a new way of working," then was pressed by commenters on X and admitted the company had simply overhired during the pandemic.

"Essentially, every large company is overstaffed. It's at least overstaffed by 25%. I think most large companies are overstaffed by 50%. I think a lot of them are overstaffed by 75%. Now they all have the silver bullet excuse: Ah, it's AI." — Marc Andreessen

This is notable because it comes from a top-tier VC whose own portfolio benefits from the AI narrative. When bulls call out the bull case, investors should pay attention.

Cursor's dependency on Anthropic was an existential concentration risk that is now playing out.

Cursor reportedly accounted for 40–50% of Anthropic's revenue at its peak — an extraordinary customer concentration for the AI lab. But the inverse was also true: Cursor was deeply dependent on Anthropic's models. When Anthropic's own Claude Code emerged as a rival coding tool, it directly threatened Cursor's core market, illustrating that platform risk cuts both ways between AI infrastructure providers and their largest application-layer customers.

"Cursor once accounted for 40% to 50% of Anthropic's revenue, uses unpaid 'work trials' to vet hires, and is weighing a potential $60 billion sale to SpaceX after Anthropic's Claude Code became a serious rival."

China's AI chip ecosystem is quietly completing its public market formation.

While Western media focuses on U.S. chip dominance and export controls, China's "four AI chip dragons" have now all reached the public markets. Enflame Technology's $888M IPO on the STAR board completes this set — a milestone that suggests China's domestic AI chip supply chain is becoming institutionally funded and independently capitalized.

Enflame Technology's approval "mak[es] it the last of China's 'four AI chip dragons' to reach the public markets."

3. Companies Identified

Anthropic | AI safety-focused AI lab | Center of the week's biggest story: forced to suspend its Fable and Mythos models within 90 minutes after a Trump administration export control order, sending senior staff to Washington to resolve the crisis. Also notable for its Claude Code product now competing directly with Cursor, its former largest revenue customer.

"Anthropic was given 90 minutes to comply with Trump administration export controls on its Fable and Mythos models, prompting the company to suspend both systems for all users."

SpaceX | Elon Musk's space and AI company | Completed a record IPO that grew to $85.7B after underwriters exercised full overallotment; shares jumped 20% on day one. Also reportedly in talks to acquire Cursor for $60B.

"SpaceX shares jumped 20% on their first full day of trading after the company's record IPO, closing at $192.50 and extending gains after Elon Musk said the company 'might be able to reach approximately' $1 trillion in revenue in 2030."

Cursor | AI coding tool (Anysphere) | Profiled as one of AI's fastest-growing startups, but now under competitive pressure from Anthropic's Claude Code. Weighing a $60B sale to SpaceX.

"Cursor once accounted for 40% to 50% of Anthropic's revenue, uses unpaid 'work trials' to vet hires, and is weighing a potential $60 billion sale to SpaceX after Anthropic's Claude Code became a serious rival."

Cerebras Systems | AI chipmaker | Debuted on Nasdaq up 68%, reaching a ~$67B market cap — the largest U.S. tech IPO since Snowflake in 2020. Shares have since fallen 30%.

"AI chipmaker Cerebras Systems closed its first day on the Nasdaq up 68% from its $185 IPO price, giving the chipmaker a market cap of roughly $67 billion — the largest U.S. tech IPO since Snowflake's 2020 debut."

Arcade.dev | AI agent access control | Raised $60M Series A for software governing which enterprise apps and databases AI agents can access or modify.

"Develops software for controlling which enterprise apps, databases, and tools AI agents are authorized to access or modify."

NewCore | Enterprise identity management for humans and AI agents | Raised $66M at $300M valuation to manage identities, permissions, and access lifecycles across enterprise systems.

"Manages identities, permissions, and access lifecycles for human employees and AI agents across enterprise systems."

Radical Numerics | AI for biology | Raised a $50M seed round for multimodal models that read DNA, RNA, and proteins to design biological systems and detect pathogens for healthcare and biosecurity.

"Develops multimodal models that read, write, and analyze DNA, RNA, and proteins to design biological systems and detect pathogens for healthcare and biosecurity."

Sarvam AI | Sovereign and multilingual AI | Raised $234M first close of a $300M Series B at $1.5B valuation, focused on generative AI for coding, cybersecurity, and sovereign AI — with HCLTech taking a 10.5% strategic stake.

"Develops generative AI models and language tools for coding, cybersecurity, and sovereign AI applications."

Salesforce | Enterprise software | Acquiring Fin (formerly Intercom) for ~$3.6B to strengthen its Agentforce AI platform for customer service.

"Acquiring Fin...for about $3.6 billion as it races to strengthen its Agentforce platform."

1Password | Password and identity security | Acquiring Apono for $250–300M, signaling that just-in-time access management for cloud infrastructure is now table-stakes for enterprise identity security vendors.

"Purchasing Apono, a four-year-old Israeli cybersecurity startup whose AI-powered platform manages just-in-time access permissions across cloud infrastructure, enterprise applications, and corporate databases."

SailPoint | Identity security | Acquired Entro Security for ~$200M to expand into non-human identity management (API keys, tokens, service accounts, machine credentials).

"Bought Entro Security...that maps, manages, and protects non-human identities such as API keys, tokens, service accounts, and machine credentials."

Block | Payments/fintech | Used as the primary case study for companies blaming AI for layoffs that were actually driven by pandemic overhiring. Jack Dorsey laid off nearly half the company.

"Pressed by commenters on X about the bloat he'd created during the pandemic, Dorsey later acknowledged that Block had, in fact, overhired."

Enflame Technology | Chinese AI chip company | Backed by Tencent; completed China's AI chip IPO cohort by winning approval to raise $888M on Shanghai's STAR board.

"The last of China's 'four AI chip dragons' to reach the public markets."

Podium Automation | Industrial control panel automation | Raised $18M Series A (a16z participating) to use software to design electrical panels, select components, and shorten manufacturing delivery times.

"Makes industrial control panels for manufacturers, using software to design panels, select electrical components, generate assembly instructions, and shorten delivery times."

Fox / Roku | Media consolidation | $25B deal combining streaming content and the largest connected-TV platform, cited as evidence of media companies racing for ad-supported scale.

"Media companies race for scale in ad-supported streaming."

Raylu | AI deal sourcing for VCs (sponsor) | Claims 4x reply rates on automated founder outreach; trusted by 50+ investment funds.

"Raylu's AI agents find companies matching your thesis, score them against your firm's investment criteria, sync bi-directionally with your CRM, and run automated founder outbound that hits 4x reply rates."

4. People Identified

Michael Truell | CEO of Cursor, age 25, MIT-trained | Profiled as the leader of one of AI's fastest-growing startups; reportedly went years without paying himself and uses unpaid work trials to vet hires. Now navigating competitive pressure from Anthropic and a potential $60B SpaceX acquisition.

"The 25-year-old MIT-trained coder who reportedly went years without paying himself while building one of AI's fastest-growing startups."

Marc Andreessen | Co-founder, Andreessen Horowitz | Identified as one of the most prominent voices pushing back on the AI-layoff narrative, calling it a "silver bullet excuse" for mismanagement.

"Andreessen said, 'Essentially, every large company is overstaffed...Now they all have the silver bullet excuse: Ah, it's AI.'"

Jack Dorsey | CEO, Block | Case study in executive messaging around AI-driven layoffs; initially attributed mass layoffs to AI-enabled productivity, then admitted to pandemic overhiring when pressed.

"Pressed by commenters on X about the bloat he'd created during the pandemic, Dorsey later acknowledged that Block had, in fact, overhired."

Andrew Feldman & Sean Lie | Co-founders, Cerebras Systems | Became billionaires on Cerebras' IPO day after shares jumped 68%.

"By the close, co-founders Andrew Feldman and Sean Lie were billionaires."

Patrick Collison | CEO, Stripe | Noted as a prior investor in Radical Numerics (the $50M AI biology seed round) and as part of the billionaire coalition fighting California's proposed wealth tax.

"Previous investor Patrick Collison also investing [in Radical Numerics]."

Charlie Javice | Founder, Frank (student-aid startup) | Convicted of defrauding JPMorgan Chase in its $175M acquisition of Frank; now seeking a presidential pardon under Trump's anticipated wave of white-collar clemency.

"Charlie Javice has been seeking a presidential pardon after being convicted of defrauding JPMorgan Chase in its $175 million purchase of her student-aid startup Frank."

Sergey Brin, Garry Tan, Ron Conway, Chris Larsen | Tech billionaires and VCs | Organized private chats, PACs, and counter-campaigns against California's proposed 5% billionaire wealth tax — but the measure still gathered enough signatures for the November ballot.

"California tech billionaires and venture capitalists...organized private chats, PACs, counter-initiatives, and candidate recruitment efforts to fight a proposed 5% tax on billionaires' net worth, only to see the measure gather enough signatures for the November ballot."

Elon Musk | CEO, SpaceX and xAI | SpaceX's record IPO swelled to $85.7B; proceeds tied to paying down X and xAI debt and expanding AI compute infrastructure. xAI also suffered its second consecutive legal loss against OpenAI.

"Elon Musk said the company 'might be able to reach approximately' $1 trillion in revenue in 2030."

5. Operating Insights

Don't let a single customer represent 40–50% of your revenue — even if that customer is a flagship AI lab.

Cursor's situation illustrates a classic but newly acute platform risk: when your largest revenue source is also a potential competitor, concentration becomes an existential threat. Anthropic's Claude Code, built on the same underlying model Cursor depends on, is now a direct rival to Cursor's core product.

"Cursor once accounted for 40% to 50% of Anthropic's revenue...after Anthropic's Claude Code became a serious rival."

Use unpaid "work trials" as a hiring filter if you're building in a capital-efficient environment.

Cursor's approach — which the article notes without criticism — is to vet hires through unpaid work trials before making offers. For early-stage founders trying to maintain lean teams under competitive pressure, this signals a growing norm of proof-of-work hiring in AI-native companies.

Cursor "uses unpaid 'work trials' to vet hires."

Build government-compliance contingency plans before you need them — not after.

Anthropic's 90-minute ultimatum and forced suspension of two major product lines is a warning for any AI company operating in dual-use or export-sensitive territory. The absence of a pre-negotiated compliance framework turned a regulatory encounter into a product crisis.

"Anthropic was given 90 minutes to comply with Trump administration export controls...prompting the company to suspend both systems for all users and fly senior technical staff to Washington."

6. Overlooked Insights

The Trump administration is quietly deregulating federal data center oversight at the exact moment AI infrastructure demand is surging.

The administration plans to let a key federal data center rule expire in September with no replacement — removing energy, water-use, security, and reporting requirements for government data centers. This is largely absent from mainstream AI policy coverage but could significantly shape where and how federal AI infrastructure gets built, with downstream implications for energy, water, and cybersecurity vendors serving the government.

"The Trump administration is planning to let a key federal data-center rule expire in September without a replacement, a move that could remove energy, water-use, security, and reporting requirements for government-run or leased data centers just as federal AI infrastructure demand is surging."

HCLTech's $150M strategic stake in Sarvam AI signals that large Indian IT services firms are actively acquiring AI capability — not just partnering with it.

HCLTech co-led Sarvam's $234M Series B and took a 10.5% stake for $150.7M. This is a significant move by a legacy IT services giant to embed sovereign AI capability at the equity level, suggesting the Indian IT outsourcing model is undergoing a structural shift toward AI-native services — a trend worth watching for its implications on global enterprise IT spending.

"The deal was co-led by HCLTech (which invested $150.7 million for a 10.5% stake) and Bessemer Venture Partners."