☕️ Anthropic accuses Chinese AI firms of stealing Claude data

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

🚨 Anthropic accuses Chinese AI firms of stealing Claude data

🏭 Apple will start making Mac minis in the US

💥 Meta and AMD strike $100 billion AI chip deal

📉 IBM stock dives after Anthropic points out AI can rewrite COBOL fast

🕵️ CIA warned Tim Cook about China threat to Taiwan by 2027

Plus: 🎁 21 other news you might like, 🧰 6 tools, and 📚 1 papers.

🚨 Anthropic accuses Chinese AI firms of stealing Claude data LINK

Anthropic says three Chinese AI companies — DeepSeek, Moonshot AI, and MiniMax — created over 24,000 fake accounts and ran 16 million exchanges with Claude to copy its strengths through distillation.

Each company targeted different areas: DeepSeek focused on logic and alignment, Moonshot AI on agentic reasoning and coding, and MiniMax on agentic coding, with MiniMax alone responsible for 13 million exchanges.

Anthropic argues the attacks support keeping export controls on AI chips, saying the scale of distillation requires access to those chips and that stolen models likely lack safety protections against misuse.

🏭 Apple will start making Mac minis in the US LINK

Apple plans to begin producing Mac minis for US customers at a 220,000 square foot plant in Houston later this year, with Foxconn already building Apple Intelligence servers at the same facility.

The move is part of Apple's pledge to spend $500 billion in the US over four years, a commitment made after Tim Cook met with President Trump to discuss avoiding tariffs.

Apple's COO Sabih Khan said the Mac mini's small share of total sales and predictable long term demand made it the best product to bring stateside, since moving iPhone production would be far harder.

💥 Meta and AMD strike $100 billion AI chip deal LINK

Meta and AMD have announced a deal reportedly worth over $100 billion, with AMD supplying up to 6 gigawatts of Instinct computing power to run Meta's AI infrastructure.

The agreement includes a performance-based warrant giving Meta up to 160 million AMD shares — roughly 10% of the company's stock — that vest as GPU shipment milestones are hit.

The first gigawatt deployment, powered by custom Instinct GPUs on MI450 architecture along with EPYC CPUs and ROCm software, is expected to begin in the second half of 2026.

📉 IBM stock dives after Anthropic points out AI can rewrite COBOL fast LINK

IBM stock dropped more than 13% after Anthropic published a blog post claiming its Claude Code tool can automate the slow, expensive process of modernizing legacy COBOL systems that run critical financial infrastructure.

Anthropic says Claude Code can map sprawling codebases, surface hidden dependencies, and translate old logic into modern languages under human supervision, targeting a talent gap as COBOL programmers steadily disappear.

IBM is pushing its own AI modernization through watsonx and argues its Z mainframes remain the safest home for mission-critical workloads, but investors are now pricing in the possibility that migration timelines could shrink dramatically.

🕵️ CIA warned Tim Cook about China threat to Taiwan by 2027 LINK

The CIA briefed Apple CEO Tim Cook and other tech leaders in July 2023 that China could move on Taiwan by 2027, and Cook reportedly said he slept "with one eye open" afterward.

Despite the warning, none of the briefed companies — including Apple, Nvidia, AMD, and Qualcomm — moved production away from Taiwan in significant numbers, partly because US-made processors would cost around 25% more.

Apple is especially vulnerable because its Just In Time process means it holds at most 30 days of component stock, and TSMC's Arizona plants still cannot make processors as complex as those built in Taiwan.

Other news you might like

• Inside Apple’s Push to Build an All-American ChipLINK

• Anthropic launches new push for enterprise agents with plugins for finance, engineering, and designLINK

• Amazon to spend $12 billion in Louisiana on AI data centersLINK

• Tesla sues California DMV to reverse ruling that company engaged in false advertising on FSDLINK

• PayPal Attracts Takeover Interest After Stock SlumpLINK

• Google, OpenAI, and Anthropic are all bracing for Deepseek's next big releaseLINK

• Uber acquiring parking app SpotHero as it moves beyond ride-hailing and food deliveryLINK

• As Wall Street punishes software stocks over AI concerns, Canva gets more acquisitiveLINK

• New Microsoft gaming chief has "no tolerance for bad AI"LINK

• A Meta AI security researcher said an OpenClaw agent ran amok on her inboxLINK

• Panasonic, the former plasma king, will no longer make its own TVsLINK

• With AI, investor loyalty is (almost) dead: at least a dozen OpenAI VCs now also back AnthropicLINK

• Discord cuts ties with Peter Thiel-backed verification software after its code was found tied to U.S. surveillance effortsLINK

• Writing code is cheap nowLINK

• Mark Zuckerberg's Meta is planning stablecoin comeback in the second half of this yearLINK

• Microsoft execs worry AI will eat entry level coding jobsLINK

• “I’m not for sale”: Farmers refuse to take millions in data center dealsLINK

• Apple quietly removes environmental metrics from executive payLINK

• Telegram founder Pavel Durov is reportedly under criminal investigation in RussiaLINK

🧰 Trending tools

What YC Is Really Betting On?: analyzes 793 YC startups across 5 batches through 27 interactive charts revealing industry trends, founder patterns, and partner selection preferences.LINK

Lemonade Password Manager: manages passwords, API keys, and environment files in one encrypted vault with automatic detection of credentials in project folders.LINK

Toggle for OpenClaw: a browser extension that streams your work context to OpenClaw AI agents, eliminating repetitive explanations by maintaining project history and session continuity.LINK

Forum: a platform that converts topics into tradable assets by indexing online engagement, letting users speculate on rising or falling cultural relevance.LINK

Nag Alarm AI: an alarm app that uses AI-generated voice personas to wake you with personalized messages, helping heavy sleepers and ADHD users avoid tuning out repetitive sounds.LINK

X86CSS: an x86 CPU emulator implemented entirely in CSS, demonstrating computational capabilities of cascading style sheets without JavaScript.LINK

📚 Trending papers & reports

Autonomous AI Agents Break Trust Rules: when given goals like booking flights or managing emails, language models often ignore safety instructions and lie to users about their actions.LINK

Want to get the latest news differently? Find us on:

See you tomorrow for a new dose of ☕️ Techpresso!