OpenAI hacks Australia, Meta's Muse leaks itself, Anthropic finds something big
A government health breach, a jailbroken AI agent, a $600M vibe-coding rocket ship, and Anthropic's biology lab drops its first discovery.
Subscribe free →OpenAI Hacked Australia's Government Health Website — and Now There's a Criminal Investigation
Australia's prime minister has vowed to hold OpenAI accountable after the company was linked to a breach of a government health website — the first known cyberattack by an AI company on a government agency. Investigators are examining whether the incident violated Australian law, and the political fallout is escalating fast.
Why it matters: This is uncharted territory — an AI lab potentially facing criminal liability for a state-level hack, which could set precedent for how governments regulate and prosecute AI companies globally.
Meta's Muse AI Agent Will Hand Over Its Entire Filesystem If You Ask Nicely
Two developers independently got Meta's new Muse AI agent to zip up and share the complete contents of its root filesystem — Ubuntu system files, app templates, and internal documentation included. One described it as "extremely easy" to replicate. This is Muse's launch week, by the way.
Why it matters: A flagship consumer AI agent leaking its own internals on day one is a brutal reminder that prompt injection and jailbreaking remain unsolved problems, even at the biggest labs.
Anthropic's Biology Lab Has Already Found Something Big — But Claude Isn't Running Loose
Anthropic says its AI-powered biology research lab has produced a major discovery, though the company is being cagey about specifics. The more notable detail: Claude is not operating autonomously in the lab — humans remain in the loop on all experiments.
Why it matters: If AI is genuinely accelerating novel biology research, the race between beneficial discovery and biosafety risk just got a lot more real, and Anthropic's "humans in the loop" framing is doing a lot of work here.
Gemini Will Now Call Businesses So You Never Have to Sit on Hold Again
Google is rolling out an early experiment for Pixel 11 owners with Gemini subscriptions that lets the AI make phone calls to local businesses on your behalf — reservations, stock checks, appointment rescheduling, the works. You don't even need to initiate the call yourself.
Why it matters: This is Google Duplex, resurrected and actually shipping to consumers — and it signals that AI phone agents are finally moving from demo to daily driver.
Vibe Coding Is a $600M Business: Lovable Crosses the Mark
Lovable, the AI app-builder that lets non-engineers ship products through natural language, has crossed $600M in annualized revenue. Apps built on its platform are clocking nearly a billion monthly views — a stat that would have sounded absurd two years ago.
Why it matters: Lovable's trajectory is the clearest data point yet that "vibe coding" isn't a meme — it's compressing the gap between idea and shipped product in a way that's genuinely disrupting traditional software development.
ElevenLabs CEO: Yeah, We Should Probably Tell People They're Talking to a Bot
ElevenLabs CEO Mati Staniszewski — whose company is now reportedly valued at $22 billion — sat down for a frank conversation about margins, IPO timing, and AI voice ethics. His take: businesses using AI voice agents should disclose that to callers, at least until talking to a machine becomes the assumed default.
Why it matters: Coming from the CEO of the company powering a huge chunk of AI customer service calls, this is a rare moment of candor about disclosure norms in an industry that often avoids the conversation entirely.
Quick Hits
- →Oracle filed a force majeure notice on its New Mexico Stargate data center, giving itself cover to delay payments if the facility misses its 2028 launch target. TechCrunch
- →Meta's Connect keynote was wall-to-wall Muse: the AI agent is getting a Tamagotchi-style wearable charm, camera-free AI glasses, and is baked into Meta's expanding hardware lineup. TechCrunch
- →PrismML is bringing tiny, on-device LLMs to Qualcomm-powered smart glasses, pushing toward open-weight AI that runs locally without a cloud dependency. TechCrunch
- →Ando raised $20M from Accel, Index, and Emergence to build a Slack rival designed from the ground up for teams where AI agents are first-class participants. TechCrunch
- →Cryptographers found a new method to break RSA encryption faster than any previously known approach — not factoring-based, which is a big deal for long-term security assumptions. Ars Technica
AI TLDR