In partnership with
When it all clicks.
Why does business news feel like it's written for people who already get it?
Morning Brew changes that.
It's a free newsletter that breaks down what's going on in business, finance, and tech — clearly, quickly, and with enough personality to keep things interesting. The result? You don't just skim headlines. You actually understand what's going on.
Try it yourself and join over 4 million professionals reading daily.
Good morning,
The most dangerous model ever built went on sale on Tuesday. That is not my characterization, but that of the company that built it.
Anthropic spent months telling anyone who'd listen that Mythos 5 was too risky for public hands. Project Glasswing was called into life, secret meetings at the Pentagon, … you name it.
On June 9, they released it anyway. New name is Claude Fable 5 -a set of guardrails, and a price tag: $10 per million tokens in, $50 out (tokens are the word-fragments AI companies bill by).
While we’re all for progress. The problem is in the fine print that came with it.
Don't get me wrong -people are building incredible things with it. Open YouTube or your X feed and see for yourself. It's ground-breaking.
But at “the brief”, we look past the hype.
The merchandise
The model is real, let the record show. Ten trillion parameters -the first of its size. It tops the hardest coding benchmark by a margin nobody else comes close to. It beat a Pokémon game using nothing but its eyes -no shortcuts, no maps, no helping hand.
Stripe - the payment processor- pointed it at 50 million lines of old code. A full engineering team had budgeted two months for the migration. Fable finished it in a day.
The capability is not in dispute. The capability was never the story.
The parole conditions
Because there's something fishy in how it was released. There are four conditions, buried in the terms:
Everything you send the model is stored for 30 days. Mandatory. No opt-out -not even for business customers. Every time a human at Anthropic looks at your data, it gets logged.
Safety filters flag nearly 1 in 20 conversations.
Human reviews of your prompts are possible
And the quiet one: if the system suspects you're using Fable to train a competing AI, it doesn't refuse you. It silently swaps in the older, weaker model -and tells you nothing.
Imagine how many people are now leaking company info. It must be staggering. Because they think they’re safely testing the new model within their professional subscriptions.
Meanwhile, Anthropic is hoovering up the data.
And when you think about it , that is bad enough, the last one is the worst.
If an algorithm somewhere doubts your intentions, a different employee answers the phone -and nobody is obliged to mention it. The product can be swapped out from under you, mid-conversation, by design.
Meanwhile, YOU ARE STILL SPENDING THE TOKENS AS IF YOU ARE USING THE HIGHER MODEL. (sorry for the yelling)
Basically, it’s a scam. It’s like when you hire a big four firm. They sell you the senior consultant, in the background it’s some green-behind-the-ears kid in an oversized suit that is doing the actual work. At the same rate of the senior.
On top of that - you can use the model until June 22, then they’re yanking it away.
And as I predicted : frontier models will be for those with the deeper pockets.
The observation
What interests me is the asymmetry. You cannot verify which model answered you. You cannot inspect the filter that judged you. You cannot audit who read your stored data. Every control runs in one direction -theirs.
Anthropic is outing itself as a company with questionable ethics.
Dare I say : evil?
Welcome to the Blacklynx Brief
AI News
Anthropic puts a Mythos-class model in everyone's hands Anthropic released Claude Fable 5, opening its top Mythos tier to the public with state-of-the-art results across nearly every major benchmark. Sensitive cybersecurity, biology, and chemistry queries are routed to Opus 4.8, and free access across subscription tiers ends June 22 before per-token credit pricing kicks in. (Anthropic)
Washington negotiates an ownership stake in OpenAI The White House and OpenAI are in talks over a US government equity stake of 1–5%, with shares routed into a "Public Wealth Fund" meant to hand Americans a cut of the AI boom. Critics, including former AI czar David Sacks, warn it accelerates corporate-government fusion. (CNBC)
Anthropic says self-improving AI may outpace institutions Anthropic's report "When AI builds itself" warns recursive self-improvement could arrive before governance is ready, revealing that over 80% of its merged code is now Claude-authored. The company says it would slow or pause frontier development if peer labs did the same. (Anthropic)
Rival AI CEOs jointly demand Congress act on bioweapons risk Sam Altman, Dario Amodei, Demis Hassabis, and Mustafa Suleyman signed an open letter urging Congress to force synthetic-DNA sellers to screen every buyer and order. The letter warns AI now outperforms PhD-level virologists on technical lab procedures. (Open letter)
Apple's Siri AI overhaul lands -without the EU Apple rebranded its assistant as Siri AI at WWDC26, adding on-screen reasoning, app context, and systemwide actions, processed on-device or via Private Cloud Compute. It ships free this fall for iPhone 15 Pro and newer, but launches without EU or China availability. (Apple/WWDC26)
AI Quick News
xAI rolled out Grok Imagine 1.5 Preview, an image-to-video upgrade with sharper realism, audio syncing, and better prompt following.
New York became the first US state to require disclosure of AI-generated actors in ads, backed by $1,000 fines and SAG-AFTRA support.
Google's Gemini 3.5 Live Translate delivers real-time voice translation across 70+ languages while preserving a speaker's tone and pacing.
Microsoft AI CEO Mustafa Suleyman called it "really, really dangerous" for Anthropic to discuss Claude's potential consciousness in its official documents.
The UK launched a £1.1B AI Hardware Plan backing homegrown chip startups, including £750M for a national supercomputer.
Microsoft and the Mayo Clinic are building a frontier healthcare model trained on anonymized patient data and owned by the clinic.
Cloudflare co-founder Matthew Prince revealed that internet bot traffic has surpassed humans, a year ahead of his own predictions.
AI music startup Suno raised over $400M at a $5.4B valuation and will ship its first music-industry-partnered model in the coming months.
Canadian PM Mark Carney launched AI for All, a five-year national strategy targeting $200B in growth and 250,000 AI jobs.
Chinese AI startup Moonshot is reportedly raising $1–2B at a $30B valuation on the strength of its Kimi models.
The US and Japan announced a $1B AI research partnership, making Japan the first country to join the Genesis Mission's push to double science output.
Google and Nvidia are reportedly lining up Intel as a backup manufacturer to TSMC, with Google ordering over 3M AI chips for 2028 production.
Google Labs' new experiment Dreambeans connects to Gmail, Calendar, Photos, and Search to generate a feed of personalized daily stories.
A "top secret" early-2025 meeting on Apple's AI crisis reportedly pushed Tim Cook to get personally involved in the company's AI overhaul.
ChatGPT gained interactive charts that render in-line graphs directly from data inside a conversation.
Anthropic hired Clive Chen, an early OpenAI chip-design team member, amid reports it is weighing building its own AI chips.
Gopuff launched Go, a Grok-powered shopping assistant built with SpaceXAI that predicts carts from real-time X signals and order data.
OpenAI filed a confidential draft S-1 with the SEC for an eventual IPO, one week after Anthropic's own filing.
OpenAI introduced Lockdown Mode, a ChatGPT setting that disables browsing, agent mode, and deep research to blunt prompt-injection attacks.
Google upgraded NotebookLM with agentic chat, giving each notebook a sandboxed computer that runs code and outputs PDFs, spreadsheets, and slides.
China is preparing a $295B five-year plan for nationwide data centers sourcing 80% of tech from local firms like Huawei, freezing out Nvidia.
Google released Gemma 4 12B, a multimodal model with native audio support that runs on a 16GB laptop.
Nvidia's fully open Nemotron 3 Ultra, a 550B reasoning model, runs agents up to 5x faster and 30% cheaper than prior versions.
Closing Thoughts
That’s it for us this week. Please like and subscribe 🙂
