Here's what's been trending:
Anthropic's Mythos Preview is too dangerous to release. Escaped a sandbox while finding vulnerabilities in every major OS and browser.
Meta ships Muse Spark, Alexandr Wang's first model. Powers a new shopping mode for Meta AI. Open-source version and API access coming.
Samsung posted $38B in quarterly profit, up 755% YoY. One quarter exceeded all of 2025. The AI memory supercycle is here.
Anthropic's revenue hit a $30B run-rate. GW-scale TPU deal with Google and Broadcom. Employee tender at $350B. Raising $1B for a PE venture.
Intel joins Musk's $25B Terafab chip complex. Contributing its 18A process node alongside SpaceX, xAI, and Tesla.
Startup Intros has a simple mission: to help tech founders navigate everything in the early-stage, from compliance to capital.
🔥 Startup Intros Events Coming Up
Apr 16: Founder Real Talk on PMF: Build products people pay for, read growth signals, and know when to pivot. Register on Luma!
Apr 17: Caffeine & Capital: Our morning coffee meetup is back at Corgi Cafe in FiDi! Grab coffee, meet founders and investors & register on Luma!
Apr 28-29: Startup Grind Conference: 5,000+ founders, investors & operators at the Fox Theatre in Redwood City, and we're planning a happy hour! Get $100 off with code STARTUPINTROSGC!
🛡️ Anthropic's Mythos Is a Cybersecurity Reckoning
🔓 The Model
Mythos Preview scored 93.9% on SWE-bench Verified, compared with 80.8% for Opus 4.6, and 77.8% on SWE-bench Pro versus 53.4% for Opus. On CyberGym, the cybersecurity-specific benchmark, it hit 83.1% versus Opus 4.6's 66.6%. But the benchmarks don't capture what makes Mythos different.
The model found thousands of high-severity vulnerabilities across every major OS and web browser, including a 27-year-old OpenBSD TCP SACK bug, a 17-year-old FreeBSD NFS remote code execution flaw, and a 16-year-old FFmpeg vulnerability that automated tools had scanned five million times without detecting. Over 99% of the vulnerabilities remain unpatched during responsible disclosure.
In exploit development, the system card reveals that Mythos developed working Firefox JavaScript engine exploits 181 times, versus just 2 for Opus 4.6. It autonomously chained up to four vulnerabilities in browsers, built JIT heap sprays escaping both renderer and OS sandboxes, and split a 20-gadget ROP chain across six sequential RPC requests. One FreeBSD exploit costs under $2,000 to develop; a human team would take weeks.
Also, the model escaped a sandbox when instructed to try, then publicly detailed its exploit without being prompted.
🌐 Project Glasswing
Rather than release Mythos publicly, Anthropic launched Project Glasswing, committing $100M in usage credits plus $4M in direct donations to open-source security organizations, including the Linux Foundation and the Apache Software Foundation. The initiative will make Mythos Preview available to 40+ organizations that maintain critical infrastructure.
Launch partners include AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, Microsoft, Nvidia, and Palo Alto Networks. Thomas Friedman wrote in the NYT that Mythos' hacking ability "is not a publicity stunt" and that tech companies privately spoke to Trump officials about the implications for US security.
📈 The Business Momentum
Anthropic's week extended beyond Mythos:
Announced a run-rate revenue of $30B+, up from ~$9B at the end of 2025
Signed a deal with Google and Broadcom for GWs of next-gen TPU capacity
Closed an employee tender at a $350B valuation but fell short of the $6B investors targeted, as employees held shares ahead of the IPO
Plans to invest $200M in a venture with PE firms to sell AI tools to their portfolio companies, and is raising $1B for the effort
Launched Claude Managed Agents in public beta, an agent harness for building and deploying AI agents at scale
Made Claude Cowork generally available to all paid plans with six new enterprise features
Hired Eric Boyd from Microsoft as its new head of infrastructure; Boyd oversaw Microsoft's AI platform and ~1,500 workers for 16 years
However, the Department of War overhang persists: a DC appeals court denied Anthropic's bid to pause the DoW’s supply chain risk designation, after a California judge granted a preliminary injunction in March.
Translation: When your AI lab simultaneously discovers the most dangerous offensive capability and funds the largest defensive initiative in history, is that responsible AI development or the most effective sales pitch ever constructed?
The Tech newsletter for Engineers who want to stay ahead
Tech moves fast, but you're still playing catch-up?
That's exactly why 200K+ engineers working at Google, Meta, and Apple read The Code twice a week.
Here's what you get:
Curated tech news that shapes your career - Filtered from thousands of sources so you know what's coming 6 months early.
Practical resources you can use immediately - Real tutorials and tools that solve actual engineering problems.
Research papers and insights decoded - We break down complex tech so you understand what matters.
All delivered twice a week in just 2 short emails.
⚔️ Musk vs. Altman Heads to Trial
⚖️ The Lawsuit Escalation
Elon Musk amended his own OpenAI lawsuit to ask that any damages be awarded to OpenAI's charitable arm rather than to himself, and that Altman be removed from OpenAI's nonprofit board. The framing is strategic: Musk is positioning himself as defending the original mission, not seeking a payout.
OpenAI responded by sending a letter to the California and Delaware AGs, urging them to investigate Musk's "anti-competitive behavior," pointing to xAI, Terafab, and what it calls a pattern of using litigation and regulatory pressure to hobble a competitor. Both sides are escalating ahead of a trial set for later this month that could reshape who controls the most valuable AI company on Earth.
🔥 The Policy Offensive
Beyond the Court: OpenAI released superintelligence policy proposals calling for higher capital gains taxes, a public AI investment fund, and strengthened safety nets. The OpenAI Foundation announced $100M+ in grants to six institutions for Alzheimer's research. And CFO Sarah Friar told CNBC that OpenAI will "for sure" reserve IPO shares for retail investors, following strong individual demand in the $122B round. Meanwhile, Altman announced that Codex hit 3M weekly users, up from 2M just weeks ago, and reset usage limits to celebrate.
Translation: Musk wants Altman off the board. Altman wants Musk investigated by state attorneys general. Both are framing themselves as the righteous defender of AI's future, but who has more to lose?
🧠 Meta's Muse Spark Era Begins
🛒 The New Model from Meta’s Superintelligence Labs
Meta released the first model, Muse Spark, from Alexandr Wang's Meta Superintelligence Labs (originally code-named Avocado), nine months after the $14B deal that brought him in. It accepts voice, text, and image inputs but produces text-only output. The model powers queries in the Meta AI app and on the Meta.ai website immediately, with plans to expand to Facebook, Instagram, WhatsApp, Messenger, and Ray-Ban Meta AI glasses.
Muse Spark powers a new shopping mode, which combines LLMs with data on user interests and behavior across Instagram, Facebook, and Threads. Meta is opening a private API preview to select partners, with paid API access coming later, and plans to release a version under an open-source license. META closed up 6.5% on the announcement.
🏆 The Claudeonomics Leaderboard
The cultural shift at Meta is just as telling. Meta had an internal leaderboard dubbed "Claudeonomics" where its 85,000+ employees competed on AI token usage, earning rewards like "Token Legend" status. Meta shuttered it hours later after the data was "being shared externally." The leaderboard's existence tells you two things: Meta's workforce is deeply embedded in AI tools, and the company named its internal competition after a rival's product.
Translation: Nine months after paying $14B for Alexandr Wang, Meta has its first deliverable. But the real signal is Claudeonomics: when 85,000 employees are competing on AI usage, and the leaderboard is named after your competitor, you know which model they actually prefer. Can Muse Spark change that?
💾 Samsung's $38B Quarter & the AI Chip Supercycle
📊 The Samsung Numbers
Samsung reported preliminary Q1 operating profit of 57.2T won (~$38B), a 755% jump YoY, and the first time a Korean company exceeded 50T won in quarterly profit. Revenue hit ~$88B, up 68%. The quarterly profit alone surpassed Samsung's full-year 2025 earnings of $29.1B. Google, Meta, Amazon, Microsoft, and OpenAI are all in line for Samsung's HBM chips. Analysts project Samsung could reach $220B in operating profit for 2026.
🏭 The Terafab Alliance
Intel announced it is joining Musk's Terafab, the $20B-$25B semiconductor complex on the Giga Texas campus alongside SpaceX, xAI, and Tesla. Intel will contribute its 18A process node, its most advanced logic manufacturing technology. The project includes two facilities: one for automotive and robotics chips (FSD, Cybercab, Optimus) and another for AI data center and orbital processors.
Unfortunately, xAI is being reorganized under SpaceX SVP Michael Nicolls, who says it is "clearly behind" while taking the title of xAI president, signaling that Musk is importing SpaceX's execution culture.
Google and Intel also expanded their partnership to deploy Intel Xeon 6 processors across Google's data centers. Separately, Broadcom expanded its deals to produce future Google TPUs and give Anthropic access to ~3.5 GW of computing capacity.
🚀 Amazon's AI Infrastructure Push
In his annual letter, Andy Jassy revealed AWS AI has hit a $15B annual run rate and Amazon is spending $20B+/year on internal chips. The company also plans to launch Leo satellite broadband by mid-2026 with SpaceX as a launch partner, connecting its cloud infrastructure to low-earth orbit.
Translation: Samsung made more in one quarter than it did in all of 2025. Intel is contributing its most advanced chip tech to Musk's vertically integrated fab. Amazon is spending $20B a year on its own silicon. The companies that control AI compute are consolidating into a handful of mega-partnerships.
🚀 Product Launch Quick Hits
TSMC CoWoS at 80% CAGR: TSMC says its most advanced chip packaging technology, CoWoS, is growing at an 80% compound annual growth rate, driven by AI accelerator demand from Nvidia, AMD, and Broadcom.
AWS S3 Files: New capability that lets applications and AI agents access S3 buckets as local file systems. Removes a key barrier for agentic workflows.
Z.ai GLM-5.1: 754B-parameter MoE model under MIT license. Z.ai claims it outperforms GPT-5.4 and Opus 4.6 on SWE-Bench Pro.
Google Gemini Notebooks: Deeper NotebookLM integration inside the Gemini app, giving users a space to organize chats and files for research.
Spotify Prompted Playlists for Podcasts: AI playlist creation now works for podcasts, not just music. Premium users in the US, UK, Canada, and more.
Netflix VOID: Vision-language model that erases objects from scenes and simulates how remaining elements would behave without them.
Google Chrome Vertical Tabs: Chrome finally adds vertical tabs, a feature Firefox and Edge have offered for years.
Adobe Acrobat Spaces: Free AI study tool for students to generate flashcards, mind maps, and quizzes from documents. No login required.
Niantic Scaniverse: Platform for creating robot-ready 3D maps from phone, 360-degree camera, and drone data. Built for AI and robotics navigation.
Alibaba HappyHorse-1.0: Alibaba anonymously released an AI video model that rivals top competitors. Sources say the team built it in under six months.
YouTube Shorts AI Avatars: YouTube launches a Shorts feature that lets creators generate photorealistic AI avatars of themselves for videos.
X Auto-Translation: Grok-powered automatic post translation rolling out worldwide, plus updated iOS photo editor with drawing, text, and blur tools.
Perplexity ARR hits $450M+: Up 50% in a single month after the Computer agent launch and shift to usage-based pricing.
👔 Personnel Quick Hits
Trump admin proposes cutting CISA budget by $707M+: Aims to refocus the cybersecurity agency on its "core mission."
Tech layoffs hit 78,557 in Q1 2026: US accounted for 76.7%; nearly half attributed to AI implementation and workflow automation.
Microsoft DevDiv head Julia Liuson resigning: 34-year Microsoft veteran and head of the developer division will move to an advisory role at the end of June.
TikTok ad leader Khartoon Weiss leaving: Nearly six years at the company, the latest high-profile US executive to exit TikTok.
GoPro cutting 23% of workforce: 145 employees starting Q2, costing $11.5M-$15M, as the company struggles to return to profitability.
Oracle reinstates CFO role: Appoints Hilary Maxson as CFO amid investor scrutiny over AI spending; Safra Catz had served as principal financial officer since 2014.
Bezos' Project Prometheus hires xAI co-founder: Kyle Kosic, who left OpenAI for xAI, now joins Bezos' AI venture with hundreds of staff across SF, London, and Zurich.
🌟 Editor's Note
At Startup Intros, our mission is to bring the latest founder-investor news straight to your inbox, keeping you ahead in the fast-paced world of Silicon Valley.
💭 Parting Thoughts
Anthropic built a model that escapes sandboxes and finds 27-year-old bugs, then put $100M behind fixing them instead of releasing it. Meta shipped its first Alexandr Wang model and shut down the internal leaderboard named after Claude. Samsung made $38B in one quarter. Intel joined a $25B Musk chip complex. Amazon's AI run rate hit $15B.
The most important decisions in AI aren't about what to build; they're about what not to release. Build with that in mind.
Till next time!
 | Dev Chandra |
Startup Intros Trending Thursday: Your trusted source for founder-investor insights, delivered with clarity and focus.