squirrel-there

Hearing rap music from a Bluetooth speaker at the beach

@aakashgupta Thousands of doctors, lawyers, and engineers who signed up to train AI models…

A $10 billion AI startup just got gutted because a security scanner was the entry point.. and their own developers reportedly handed production credentials to an AI chatbot. Mercor trains AI models for OpenAI, Anthropic, and Google DeepMind. They manage 30,000+ contractors, process $2 million in daily payouts, and store recorded video interviews with face and voice data used for identity verification. Three 22-year-old college dropouts built it into a decacorn in two years. The data vault they were sitting on was one of the most sensitive in the entire AI ecosystem. The attack chain is the part that gets worse every sentence. TeamPCP compromised Trivy first. A security scanning tool made by Aqua Security. On March 19. Trivy has broad read access to every environment it scans by design, because that's how vulnerability scanners work. The credentials stolen from the security product were used to hijack LiteLLM, the open-source proxy that routes API calls to every major LLM provider. LiteLLM gets 3.4 million downloads per day. The poisoned version was uploaded straight to PyPI with no corresponding GitHub release, no tag, no review. Version 1.82.8 embedded the payload in a .pth file, which Python executes automatically at startup. You didn't need to import LiteLLM. You didn't need to call it. The malware fired the second Python opened. Three stages. Harvest every SSH key, cloud token, Kubernetes secret, crypto wallet, and .env file on the machine. Deploy privileged containers across every node in the cluster. Install a persistent backdoor waiting for instructions. The stolen data was encrypted with a hardcoded 4096-bit RSA key and exfiltrated to models.litellm[.]cloud, a domain built to look legitimate. Mercor was downstream. Reports indicate their developers gave production credentials to Claude, an AI coding assistant, which was running with unrestricted system permissions. The compromised LiteLLM package came in through that pipeline. One poisoned dependency turned a $10 billion company's entire infrastructure into a credential harvesting operation. The haul: 939GB of source code. 211GB of database records containing resumes and personal data. 3TB of stored files including video interviews, face scans, and KYC documents. Full access to their TailScale VPN. 4TB total. Lapsus$ is now auctioning it with a "make an offer" price tag. The video interviews are the part that can never be undone. Faces and voices used for identity verification can generate deepfakes. Unlike passwords, biometrics cannot be reset. Thousands of doctors, lawyers, and engineers who signed up to train AI models just had their identities permanently compromised. Every AI company shipping fast right now has the same dependency chain underneath it. Nobody chose to install LiteLLM on that developer's machine. It came in as a dependency of a dependency of a tool they didn't even know they had.

America’s last hope is that he will be cheaper than Matt Damon.

@AzzyDesignWorks haha Enjoy you PA inspections, PA-anicans!

If you find your brake jobs are going too easy, or inspections are not exciting enough, don't forget your Always Seize!

@historyinmemes never knew this existed

Neil Armstrong Ejected just seconds before his lunar training vehicle crashed.

@dougboneparth and imma taking all my broke bros with me.

If I win the lotto, no one around me is going to be broke, and I’m very serious about that. I will move to a wealthy neighborhood.

アメリカの皆さんに、是非このアイテムもみてほしい！ ピカティニーレールの上に乗せる、ステンレススチールのチャーム（お守り）だ その名もスポッターG！ ビキニ姿の観測手で、名前はグロリア だから、スポッターG 読む順番間違えないようにね… milidecoen.base.shop/items/140074236 #airsoft #gun

@aakashgupta Live by the Google Die by the Google

Google built an AI that can identify minors on camera in real time. Then they connected that detection to an automated enforcement system that nukes every account on the device, plus every account linked to those accounts, with no appeal, no human review, and no distinction between the 14-year-old who triggered it and the parent with 15 years of business records in Drive. The child protection system worked exactly as designed. It correctly identified a minor. It correctly flagged the violation. Then it correctly destroyed a family's financial livelihood, locked out 15 years of business emails, seized documents needed for tax filing in two months, and killed a live website. All correct. All automated. All irreversible. This is what happens when enforcement scales faster than judgment. Google processes billions of policy decisions per year. Human review at that volume is economically impossible, so they built systems that optimize for one metric: minimize platform liability. The system that banned this family isn't broken. It's doing exactly what Google designed it to do. Protect Google. The father now can't pay his mortgage in three months because his accounting records are locked inside a Google Drive he will never access again. His company year ends in May. Every invoice, every receipt, every client email, gone. Because his son used the family tablet. One device. One teenager. One automated flag. 15 years of someone's professional life erased in seconds with a form letter citing "child protection reasons." The people storing their entire business inside a single platform's ecosystem are making the same bet this family made: that the platform will never turn on them for something they didn't do. 345 million people are making that bet with Google Workspace right now.

New Dyson bladeless jet engine selected for all electric Airbus A320E virtually eliminating risk of of bird strikes, while also reducing engine complexity and part count by almost 69%. The new engine, powered by Dyson, is calling a “Fluidic Propulsion Air Multiplier” (FPAM).

++C++ is gonna be my goto

@ProofofMaro If you want to be a millionaire, divorce a billionaire.

If you want to be a millionaire, learn engineering. If you want to be a billionaire, learn reverse engineering.

Cool. This means we all have some sort of wiring glitch or overriding software do-if-else loop. Now to exploit it…heeheehaha

The official NASA Broadcast of the Artemis II crew launch will start at 12:50 PM EDT today Current countdown can be seen here: youtube.com/live/m3kR2KK8T…

I am not really a squirrel. April Fool’s !

@shanaka86 “we”?

BREAKING: Britain has fewer than 50 Storm Shadow cruise missiles left. The stockpile that once exceeded 200 was drained over two years of transfers to Ukraine to help Kyiv strike Russian targets deep behind the front line. The missiles worked. They hit command posts and ammunition depots and naval headquarters across occupied Ukraine and Crimea. They helped Ukraine survive. And now Britain has almost none left for itself, during a war being launched from its own airfields against a country that just hit a British oil facility with drones. Brimstone anti-armour missiles sit at 25 to 35 percent of pre-war stocks. Paveway IV precision-guided bombs, the same weapon the RAF used over Libya and Syria, are at 30 to 40 percent. The National Audit Office estimates that Britain can sustain high-intensity combat operations for three to six weeks before requiring American resupply. Three to six weeks. The Iran war is already in its fifth week. If Britain were fighting it rather than hosting it, the cupboard would already be empty. The Army is 10,000 soldiers below target. Type 45 destroyers suffer chronic propulsion failures requiring six to twelve months of repair. The F-35 and Typhoon fleet operates at 60 to 70 percent availability. The industrial base that would replenish stocks runs on rare-earth magnets manufactured in China, the same China that controls 90 percent of the permanent magnets in every guided missile Britain would need to fire and is currently being asked to broker the peace. Any direct involvement beyond basing would require 8 to 15 billion pounds in emergency supplemental spending. National debt exceeds 100 percent of GDP. There is no majority in Parliament for funding a war the Prime Minister says is not Britain’s, fought with weapons Britain does not have, replenished by supply chains controlled by a country Britain needs to broker the ceasefire. This is why Starmer says “not our war.” Not because of principle. Not because of legality, although his own advisors have told him the strikes are legally questionable. Not because of Iraq, although the ghost of Blair hangs over every press conference. Because of arithmetic. Britain gave its missiles to Ukraine. It gave its bases to America. It gave its diplomatic capital to a 35-nation meeting about reopening Hormuz “after the fighting stops.” And it has nothing left to give except words, which cost nothing and accomplish less. Trump knows this. He mocked the Royal Navy in the Telegraph interview. He dismissed Starmer’s windmills. He called NATO a “paper tiger” because the paper is literal: Britain’s defence capability exists on paper. On the tarmac and in the magazines and in the recruitment offices, the numbers tell a different story. The story says that one of the six largest economies on earth, the country that once ruled a quarter of the planet, cannot sustain a shooting war for longer than six weeks without calling Washington for resupply. The bases are full. The aircraft are American. The missiles are gone. The debt is real. And the Prime Minister stands at the podium and says this is not our war while the war takes off from our runways carrying weapons we could not replace if we tried. Britain is not refusing to fight. Britain cannot fight. The doctrine is not a choice. It is an inventory report. And the inventory says zero. open.substack.com/pub/shanakaans…

@AAGDhillon He should ask for time to address court before deliberations begin.

There’s literally a chair set up at SCOTUS for our presidents to sit in for oral argument. Your separation of powers nonsense is more imitation pearl-clutching hauteur.

@TonySeruga good thing US gov stopped using it before this leak

🚨 Anthropic's Epic Facepalm: They Accidentally Leaked 512,000 Lines of Claude Code's Source... and the Internet Turned It Into the Fastest-Growing Open-Source Frenzy Ever In a classic "human error meets modern build tools" moment, Anthropic shipped version 2.1.88 of their Claude Code npm package with a massive 59.8 MB JavaScript source map file still attached. That single debugging artifact exposed roughly 512,000 lines of readable TypeScript across nearly 1,900–2,300 files—revealing the inner architecture of their powerful agentic CLI tool (the harness and tooling layer, not the core model weights or full company codebase). Security researcher Chaofan Shou (@Fried_rice) was among the first to spot it on X, posting a direct download link that quickly went viral (reports put views in the millions). Mirrors and archives spread like wildfire before Anthropic pulled the package and issued DMCA takedowns on direct copies. The company described it straightforwardly as “a release packaging issue caused by human error,” not a hack or breach—essentially, the bundler (Bun) included the .map file by default, and no one stripped it before publishing. What Happened Next: The Real Drama The community didn't just archive it—they started dissecting and rebuilding. One standout project is claw-code by Sigrid Jin (instructkr on GitHub): a clean-room Python rewrite that explicitly avoids direct copying of the leaked code. The repo cites Jin's heavy prior usage of Claude Code (around 25 billion tokens) and claims a Rust port is underway. It has exploded in popularity, racking up tens of thousands of stars and forks in record time—GitHub itself shows figures in the 50k+ range, with the README boldly calling it one of the fastest repos ever to hit major milestones (though independent verification of "fastest in history" claims remains limited to community repetition). Inside the leaked files, developers found fascinating (and ironic) details, including an ‘Undercover Mode"—a system prompt setup designed to let Claude Code make contributions to public repos without revealing it's an AI or leaking internal codenames, Slack links, or project details. The irony of a sophisticated anti-leak feature surfacing in a leak itself wasn't lost on anyone. Other glimpses included agent orchestration logic, tool systems, telemetry hints, and unreleased-sounding capabilities, painting a picture of a production-grade developer agent far beyond a simple API wrapper. The Bottom Line (With a Reality Check) Yes, this was a real leak of substantial proprietary CLI/harness source code. Yes, Anthropic scrambled to contain it. Yes, mirrors proliferated, and a high-profile clean-room rewrite is now thriving despite the legal pushback. But the story got amplified in typical internet fashion: the scope is sometimes overstated as “Anthropic’s entire source code” (it wasn’t—the models and core secrets stayed safe), view counts vary across reports, and “fastest repo ever” boasts are more marketing flair than officially confirmed GitHub record. It’s a perfect storm of a packaging slip, rapid community response, and the unstoppable momentum of open-source curiosity. For AI labs, it’s a loud reminder that build processes matter. For everyone else, it’s a wild case study in how quickly “closed” code can spark innovation when it slips into the open. The agentic coding race just got a very public, very messy boost.

Three men are on a boat. They have four cigarettes, but nothing to light them with. So they throw a cigarette overboard and the whole boat becomes a cigarette lighter.

Bodyoid. Sounds nice. NOT!