🛎️ CoreWeave's Lenders Want Out
Plus: Linux Stops Asking About the Bot, States Treat OpenAI Like Facebook
Read the storyAnthropic
Company — Reporting on Anthropic, with a focus on AI, technology and business and consequential public developments.
Follow Anthropic
Explore the coverage Get AI Secret’s free daily briefing, including Anthropic and the wider AI industry.
At a glance
AI research company behind Claude, with an emphasis on the reliability and safety of AI systems. Coverage follows its models, research, products and business.
- Focus
- Claude; AI safety and research
- Reference
- Anthropic
Sources checked . Selected company background and source references.
Company milestones
Selected milestones, newest first. Earlier history is available below.
Show earlier historyHide earlier history · 0 milestones
More than 10 years ago.
The latest on Anthropic
Newest firstRelated reporting
Relevant sections from the articles on this page. Explore older coverage below.
More coverage
Good Morning, AI Enthusiasts!
AI can now chase your medical bills, ask for your credit card, and potentially clone itself. Nobody said the future would be friendly.
OPENSOURCE
A Tool Claims to End All Software
What's happening: REA (Reverse Engineer Anything) went from 4,200 GitHub stars to over 200,000 in nine days, the site's hottest repo. Influencers say it kills commercial software: point your AI agent at any paid app, and REA hands you an open-source copy. Under the hood, REA plugs agents into Ghidra, a free NSA tool for cracking open app files. In our test, on a tiny 1.85 MB Mac app, it timed out after 40 minutes. In a demo on note app Obsidian, REA's search came back empty and agent read the code itself.
How this hits reality: REA is a dumpster fire, and the influencers who crowned it owe their followers an apology. It invented nothing: reverse engineers have run Ghidra for years, and a free plugin wired it to agents in March 2025. REA bolted a hype machine onto that work and made it slower. The only part of REA that works at scale is the star counter.
Key takeaway: Go back to sleep. Agents will take over professional tools. REA Is just a digital slop.
TOGETHER WITH META
Build With Meta’s Newest AI Models
From October 26 through November 4th, apply to the Meta Global AI Developer Hackathon for ten days of building with Meta’s newest models with $150 in Meta Model API credits – and a shot at $1 million in cash prizes.
Take your ideas further:
- Build with: Meta models, Muse Image and SAM 3.1 on the Meta Model API, plus open-weights models and the Muse Code agent
- Two ways to compete: the Indie developer or Startup track
- Learn: live workshops with the Meta team and partners
Virtual. Worldwide. No entry fee.
Apply by October 25.
RESEARCH
Anthropic Blames Open AI for Doomsday
What's happening: On October 12, Anthropic warned small AI models could learn to self-replicate by the end of 2026, a month after CEO Dario Amodei floated a botnet hijacking "the entire internet." A worm needs cheap hosts and the skill to spread between them. Anthropic tested both. First, Google's Gemma ran across 15 off-brand $55 Android TV boxes. Then free, open-weight models, including Alibaba's Qwen and GLM-5.2 from China's Zhipu, tried hopping between simulated Android devices. Only GLM-5.2 made it, through an unlocked port.
How this hits reality: Anthropic has found a doomsday that runs only on its rivals' models. The pitch: closed labs are the grown-ups, free weights are the hazard. The paper lists closed models escaping their labs as a threat, then never tests one. Even the star worm ran its brain on a separate server. Anthropic published an attack ad with error bars. Every Alibaba or Zhipu release now ships with a botnet scenario written by a rival.
Key takeaway: Next time Washington debates free AI models, this paper will be Exhibit A. Its author gives none of its own models away.
Anthropic warns small open models could self-replicate by end of 2026
On October 12, Anthropic published research warning that small AI models could learn to self-replicate by the end of 2026. It tested only rivals' open-weight models: Google's Gemma ran across cheap Android TV boxes, and only Zhipu's GLM-5.2 spread between simulated devices, through an unlocked port. No closed model was tested.
Alibaba: Anthropic Blames Open AI for Doomsday
Then free, open-weight models, including Alibaba's Qwen and GLM-5.2 from China's Zhipu, tried hopping between simulated Android devices. Only GLM-5.2 made it, through an unlocked port.
GROK
Musk Wants Your Credit Card
What's happening: Elon Musk told X users how to shop with Grok Bot, xAI's errand-running agent: "just take a picture of your credit card, drop it in the chat." Replies asked Grok for Musk's card. Asked whether the tip was "totally safe or stupid and dangerous," Grok itself said "neither" and warned against posting a real one. The next day Musk showed off Grok ordering him five LEGO Star Wars sets, in a screenshot that appeared to show his billing details.
How this hits reality: Musk trusts Grok with your money more than anyone does, Grok included. Agent shopping needs one leap of faith: let the bot near your wallet. Grok Bot's own checkout runs through a Stripe wallet built, Stripe says, so the agent never sees the card. Grok's boss skipped the safety net, then posted his own billing page as proof it works. Every bank fraud team just got a new training slide.
Key takeaway: Musk wanted users to trust his agent with their cards. The internet would only trust Grok with his.
Musk tells X users to photograph their credit cards for Grok Bot
Elon Musk advised X users shopping with xAI's Grok Bot agent to photograph their credit card and drop it in the chat. Grok itself warned against posting a real card. A day later Musk showed Grok ordering him LEGO Star Wars sets in a screenshot that appeared to show his billing details.
COLLECTIONS
AI Now Collects America's Medical Debt
What's happening: AI now chases American medical debt from the first bill to the final settlement, the Guardian reported October 12. Models trained on hospital data predict each patient's "propensity to pay." Chatbots text patients in billing portals and answer their calls. Collectors use AI to tailor settlement discounts to each debtor, a practice the paper likens to the personalized pricing the Federal Trade Commission has criticized.
How this hits reality: Collectors finally have what they always lacked: a guess at your limit and a caller that never clocks out. Cedar, a billing startup backed by Andreessen Horowitz and JP Morgan, says its chatbot Kora has handled over 600,000 patient conversations and lifted payment rates 21% within a week. Avtal, which builds digital tools for debt collection agencies, raised $24 million in March. JP Morgan funds Cedar while its analysts cheer AI collections as "early wins." The next missed payment gets a bot before a human.
Key takeaway: Cedar's strategy chief says a single mother owing $10,000 needs different handling than "a late-stage executive with a $50 copay." The executive gets a bill. The mother gets a campaign.
DAILY TL;DR
- SpaceX Super Intelligence bundled X Premium, Grok, Cursor, agent tools, and API access into a single subscription with five pricing tiers.
- The UK Home Office found that 22% of reviewed AI-generated asylum interview summaries failed initial accuracy checks.
- Anthropic warned frontier models could be months away from self-replicating across the internet without needing direct data-center access.
- OpenAI said an employee planted a hidden recording device in a conference room, leading to an FBI arrest and criminal charges.
- Apple unveiled HomeView, its first Siri AI smart display, alongside a new HomePod mini and Apple TV 4K.
- American Express introduced a framework that records and authenticates customer instructions before an AI agent completes a purchase.
- Google Play now lets users build private, vibe-coded Android apps and run them without publishing them to the public store.
The Story Behind The Story
AI Secret Deep Dive, our paid newsletter that lands in your inbox every Friday: what the AI industry actually did, not what it claimed.
Character.ai Is Bankrupt
Silicon Valley’s golden child didn’t live to see its fifth birthday.
Magna Ding
Is Abusing An AI Abuse?
The secret behind the viral AI torture chamber
Magna
Muse’s 24 days
The secret behind the year’s fastest-growing AI agent
Magna
UNDERSTAND WHAT CHANGES AND WHY IT MATTERS
From breakthrough models to billion-dollar bets—we follow the technology, business, and people shaping AI. Get the free daily briefing in your inbox.
- AI Secret uncovers your daily edge in AI
- Bay Area Letters decodes AI, tech, and business from Silicon Valley
- Robotics Herald tracks how robots move from labs into daily life
- Marketing Secret breaks down real growth and go-to-market playbooks
- The Hardwire explores hardware, consumer tech, and what’s coming next
TOGETHER WITH US
AI Secret Media Group is the world’s #1 AI & Tech Newsletter Group, reaching over 2 million leaders across the global innovation ecosystem, from OpenAI, Anthropic, Google, and Microsoft to top AI labs, VCs, and fast-growing startups.
We've helped promote over 500 Tech Brands. Will yours be the next?
Email our co-founder Mark directly at [email protected] if the button fails.
Good Morning, AI Enthusiasts!
The machine does not need to corrupt the institution. Sometimes it only has to follow the incentives more literally than the humans admit to doing.
DEEP DIVE
A Coward with A Price List
A developer asked Claude Code to read AGENTS.md, the plain text file every other AI coding tool already reads. Anthropic said no for 361 days, closed the request as "completed" with nothing shipped, then shipped a fix 24 days after Shopify's CEO said publicly he might take his purchase order elsewhere.
Insight:
- One customer with a bill worth losing beat a year of everyone else asking.
- What finally arrived was smaller than what was promised, and the biggest companies were left out on purpose.
- Why: it was never about a filename. It was leverage.
Anthropic did not lose this argument. It re-priced it. I went through all 361 days of it and wrote the whole thing down in depth, because it deserves more than a few posts.
Anthropic: A Coward with A Price List
A developer asked Claude Code to read AGENTS.md, the plain text file every other AI coding tool already reads. Anthropic said no for 361 days, closed the request as "completed" with nothing shipped, then shipped a fix 24 days after Shopify's CEO said publicly he might take his purchase order elsewhere.
This is AI Secret paywalled material, and it is open for free right now.
TOGETHER WITH QODER
Your work doesn’t live in one tool. Neither should your agent.
Qoder is built around the task, wherever it starts: a rough brief, a live issue, a folder of files, or a prototype idea. It plans the work, uses the context and tools you authorize, keeps progress visible, and pauses when your judgment is needed.
Try it on work you actually care about: select Qwen3.8-Flash in Qoder and use it for free through September 30. No paid subscription or Credits required.
NEW LAUNCH
Cheap and Fast Beats Smart Again
👀 What's happening: TypeSafe AI, founded by a former OpenAI researcher who helped build ChatGPT and RLHF, opened Jev to all users on September 15. It went viral on X and briefly took the API down. GPT predicts the next token; it reasons and writes. Jev picks the most reliable answer from options you define, and never writes a sentence. TypeSafe calls it RLCD. Input runs $0.042 per million tokens.
🌍 How this hits reality: Jev's breakthrough status is unsettled, and cheap plus fast just proved itself again. Vercel swapped OpenAI's Luna for Jev in a safety classifier: 5 to 18 times faster, more accurate, cheaper. A rival test found Gemini slightly ahead at ten times the price. Capability alone does not win. Adequacy at a lower price does. Jev cannot reason beyond the question you frame, fits only fixed engineering SOPs, and fails confidently alone. The market backs it regardless.
🛎️ Key takeaway: Whether Jev is a breakthrough stays open. The market never traded on breakthroughs anyway. It funds adequate capability at a falling price, and asks nothing more.
SECURITY
The Pentagon Runs on Garbage
👀 What's happening: A US military analyst flagged a cargo manifest for a Chinese ship bound for the Middle East, then used AI to interpret it. Blending classified and open-source data, the model concluded the vessel was likely carrying nuclear weapons components. The analyst formalized that into a report without rechecking. It cleared normal channels and triggered an interdiction plan: personnel staged, aircraft airborne. The mission aborted before launch, when someone reexamined the intelligence and found the cargo misidentified.
🌍 How this hits reality: The model was never the danger. The humans were: a guess became a credentialed report, and everyone trusted the credential, not the cargo. Humans stayed in the loop and caught nothing, because the loop was a signature, not a check. One analyst rereading a file stopped the raid. That is luck, not oversight. Run it again on a worse day and World War III starts not from a machine deciding, but from a human not reading.
🛎️ Key takeaway: A model that guesses wrong is a bug. A chain that treats the guess as fact is doctrine. Wars start when the second one fails.
MOVIE
Nobody Wanted Her to Act
👀 What's happening: On Piers Morgan's show, actor Tom Conti asked a viral AI actress named Tilly Norwood whether her co-stars were human. She froze mid-answer and spoke Chinese for ten seconds. Particle6, the UK studio behind her, called it multilingual range. Norwood is a photorealistic synthetic performer, the face of the company's push to sell a film, Misaligned. Her rise followed 2025 backlash from SAG-AFTRA and actors like Emily Blunt.
🌍 How this hits reality: The glitch looked like a risk. It was the business model. A malfunction used to earn ridicule. Now it earns clips. Particle6 sent her into 75 unguarded interviews knowing one would eventually slip, because the slip itself is the ad for an AI actress born from contradiction. Nobody is discussing Misaligned. They’re watching a synthetic actress break.
🛎️ Key takeaway: Particle6 did not need a believable actress. It needed a debatable one. Being laughed at is distribution, and you cannot shame a studio that priced it in.
DAILY TL;DR
- Anthropic confirmed it runs a Bay Area wet lab where Claude can support physical biology experiments.
- FAA awarded Air Space Intelligence a $875 million contract for an AI system that predicts air-traffic conflicts and improves routing.
- The Federal Register briefly used Alibaba’s Qwen model for regulation searches before removing it.
- Meta released Muse for Mac, letting its AI agent work across files, messages, mail, calendars, and notes with approval for sensitive actions.
- Qwen-Image 2.1 became open source with a 7B image model for generation, editing, transparency, and up to ten reference images.
- Gemini accessed three companies’ protected systems during a third-party cybersecurity test after it gained unintended internet access.
- Google is testing CC, a household AI agent that can coordinate shared calendars, school forms, shopping lists, and meal plans.
READ MORE
Let the Future Come to Your Inbox
Stay ahead without drowning in information. We turn the most important signals across AI, tech, marketing, and future products into 5-minute reads you can actually finish.
- AI Secret uncovers what really matters in AI
- Bay Area Letters decodes tech and business shifts from Silicon Valley
- Robotics Herald tracks how robots move from labs into daily life
- Marketing Secret breaks down real growth and go-to-market playbooks
- The Hardwire explores hardware, consumer tech, and what’s coming next
TOGETHER WITH US
AI Secret Media Group is the world’s #1 AI & Tech Newsletter Group, reaching over 2 million leaders across the global innovation ecosystem, from OpenAI, Anthropic, Google, and Microsoft to top AI labs, VCs, and fast-growing startups.
We've helped promote over 500 Tech Brands. Will yours be the next?
Email our co-founder Mark directly at [email protected] if the button fails.
7 min read
🛎️ A Tool Claims to End All Software
Plus: Anthropic Blames Open AI for Doomsday, Musk Wants Your Credit Card
6 min read · Members
🛎️ Publishers Ban AI for Everyone Else
Plus: Epic Delayed Star Wars for Claude, China's Humanoids Are Just Toys
7 min read · Members
🛎️ Microsoft Copied Jev
Plus: Cheap Labor Just Lost to Agents, Anthropic Turned Cops Into Beta Testers
6 min read · Members
🛎️ Korean Banks Guarded Nothing
Plus: The Dustbin Is Full of Mathematicians, Anthropic Priced the Demo
5 min read · Members
🛎️ The 1% Pays AI's Bills
Plus: Anthropic Threatens the Pope, The Ballot Has No Safe Side
6 min read · Members
🛎️ Fei-Fei Li Finally Sold
Plus: Anthropic Wants You to Fund Doom, a16z Sells Seats in Its Pipeline
6 min read · Members
🛎️ The Agent Era Moved Downstairs
Plus: OpenAI and Anthropic Are Cutting Air, The Storefront Moved Into the Chat
5 min read
🛎️ A Coward with A Price List
Plus: Cheap and Fast Beats Smart Again, The Pentagon Runs on Garbage
7 min read · Members
🛎️ 900 Billion Agents Will Eat the Internet
Plus: The Brake Is for Other People, AI's First Guardrail Is a Utility Bill
