💸 Anthropic commits $100B to Amazon’s AWS ↗
Anthropic and Amazon just tightened the bolts on one of the biggest AI infrastructure tie-ups around. Anthropic is committing more than $100 billion to AWS over the next decade to train and run Claude.
Amazon is putting in $5 billion immediately, with the option to invest up to $20 billion more. The deal also gives Anthropic access to Amazon’s Trainium chips - basically Amazon saying, “please don’t just rent Nvidia forever,” but with a much larger cheque.
The funny bit, or maybe not funny at all, is that Anthropic is still under regulatory pressure while becoming even more deeply plugged into Big Tech infrastructure. Very clean, very tangled.
🧾 Anthropic outspends OpenAI in record lobbying quarter ↗
Anthropic spent $1.6 million on lobbying in the first quarter, while OpenAI spent $1 million. That makes it a record quarter for both - and a pretty clear sign that AI labs are no longer just building models, they’re working Washington like seasoned operators.
Anthropic’s spend jumped sharply compared with the same period last year, reportedly tied to federal scrutiny, Pentagon friction, and the broader battle over how governments should treat frontier AI.
OpenAI’s lobbying focused on copyright, cybersecurity, AI policy, and cloud infrastructure. Not exactly glamorous stuff, but this is where the future gets quietly stapled together.
🧠 OpenAI launches ChatGPT Images 2.0 ↗
OpenAI rolled out ChatGPT Images 2.0, with better instruction-following, stronger text generation inside images, and new “thinking” abilities that can use web search to help create more complex visuals.
The upgrade is aimed at things like infographics, slides, maps, manga, multilingual text, and multi-image generation. The line between “image tool” and “tiny design department in a box” is getting suspiciously thin.
The big shift is control. OpenAI is trying to make image generation less like shaking a magic snow globe and more like directing a slightly caffeinated art intern.
🎭 YouTube expands AI deepfake detection for celebrities ↗
YouTube is expanding its likeness detection tool so more public figures can find AI-generated videos that imitate them and request removals.
The system scans for AI lookalikes and flags them to enrolled celebrities, who can then ask YouTube to review the content under its privacy policy. Not every takedown will be approved, which is where the whole thing gets chewy.
It’s another sign that deepfakes have moved from “internet panic” to actual platform plumbing. The pipes are being labelled now, finally.
⌨️ Meta will record employee keystrokes to train AI models ↗
Meta is rolling out an internal tool that captures employee mouse movements, button clicks, and keystrokes, then converts that activity into training data for AI models.
The idea is to teach AI systems how real people navigate software and complete tasks. Valuable? Yes. Slightly dystopian office jazz? Also yes, both things can be true.
This is the agent race in miniature: companies need human workflow data with all its frayed edges, not just polished text scraped from the internet. So now the keyboard becomes a witness. Bit grim, bit clever.
🛡️ Unauthorized users reportedly accessed Anthropic’s Mythos cyber tool ↗
A group of unauthorized users reportedly gained access to Anthropic’s Mythos, a cybersecurity-focused AI tool that was meant to be tightly controlled.
Mythos is not a general release product - it’s designed for advanced cyber work, which makes the access claims extra spicy. If true, it raises uncomfortable questions about how labs secure the very systems they describe as sensitive.
The whole thing has that “locked room mystery, but the room is a cloud dashboard” feeling. Very on-brand for AI security right now.
🧪 NeoCognition raises $40M seed to build human-like learning agents ↗
AI research lab NeoCognition raised a $40 million seed round to build agents that learn more like humans.
The startup is chasing more reliable and efficient AI systems - a very hot pitch as investors keep hunting for anything that might make agents less brittle in practical workflows.
It’s another reminder that the agent boom is not just OpenAI, Anthropic, and Google swinging frying pans at each other. The smaller labs are circling too, some with properly ambitious moon-rock energy.
FAQ
Why is Anthropic committing $100 billion to AWS?
Anthropic is committing more than $100 billion to AWS over the next decade to train and run Claude. The deal gives Anthropic deeper access to Amazon’s cloud infrastructure and Trainium chips. It also reflects how costly frontier AI systems have become, with model development increasingly dependent on vast compute partnerships.
What does the Anthropic and Amazon deal mean for AI infrastructure?
The Anthropic and Amazon deal shows how closely AI infrastructure is becoming tied to Big Tech cloud platforms. Amazon is investing billions, while Anthropic is committing huge long-term spending to AWS. These arrangements can help AI labs scale, but they also raise questions about dependency, competition, and regulatory scrutiny.
Why are Anthropic and OpenAI spending more on lobbying?
Anthropic and OpenAI are spending more on lobbying because AI policy is becoming central to their future. The article says Anthropic spent $1.6 million in the first quarter, while OpenAI spent $1 million. Their focus areas include federal scrutiny, copyright, cybersecurity, AI policy, cloud infrastructure, and how governments should regulate frontier AI.
What is new in ChatGPT Images 2.0?
ChatGPT Images 2.0 is described as improving instruction-following, text generation inside images, and “thinking” abilities that can use web search for more complex visuals. The upgrade is aimed at infographics, slides, maps, manga, multilingual text, and multi-image generation. The main shift is toward greater control and less guesswork in image creation.
How is YouTube handling AI deepfakes of celebrities?
YouTube is expanding its likeness detection tool so more public figures can find AI-generated videos that imitate them. Enrolled celebrities can be alerted to possible AI lookalikes and request removal reviews under YouTube’s privacy policy. Not every request will automatically be approved, so the system still depends on platform review and context.
Why is Meta recording employee keystrokes for AI training?
Meta is rolling out an internal tool that captures employee mouse movements, clicks, and keystrokes, then turns that activity into AI training data. The goal is to teach AI systems how people navigate software and complete tasks. In many agent workflows, this kind of human activity data may help models learn practical, step-by-step behavior.