Cortech Daily

A record European privacy fine over automated driver bans, a surveillance company negotiating in public, and the first malware documented inside a car's dashboard.

(0:00) - Segment 1

(0:24) - Uber faces fine of nearly $1B over automated driver suspensions - source

(1:23) - Flock CEO calls for 'compromise' as surveillance company faces growing backlash - source

(2:15) - Android car head units infected with proxy botnet malware through built-in software updaters - source

(3:03) - Ransomware attackers are zeroing in on mid-market companies - source

(3:57) - Fake bank websites play dead to evade security scanners - source

(4:36) - Who's behind the new 'stealth model' Ox Alpha? - source

(5:25) - Anthropic's best AI model struggles to attract users as cheaper tools thrive - source

(6:23) - Is it legal to train AI models on copyrighted books? It's complicated - source

(7:10) - Kids outlearn AI - and we still don't know why - source

(7:50) - Fable and the end of the free lunch - source

(8:43) - Segment 12

Sources: every feed this show reads, curated in Don't Hype Me.

Anthropic's two trillion dollar I P O path, a new Model Context Protocol roadmap, frontier labs scored on rogue-model containment, Uber's near-billion-dollar Dutch fine, and the memory shortage hitting Amazon's shelves.

(0:00) - Intro

(0:23) - Anthropic heads toward the largest I P O on record - source

(1:09) - Model Context Protocol publishes a five-point roadmap - source

(1:58) - Frontier labs score poorly on rogue-model containment - source

(2:44) - OpenAI reverses course and asks California to toughen S B fifty three - source

(3:27) - Prime Intellect benchmarks eighteen models on autonomous research - source

(4:11) - Dutch regulator fines Uber eight hundred twenty five million euros - source

(5:01) - TikTok and ByteDance settle children's privacy case for four hundred million - source

(5:47) - Connecticut says forty one thousand Medicaid members were exposed - source

(6:34) - Residential proxy networks keep growing despite takedowns - source

(7:23) - Amazon raises device prices by as much as sixty percent - source

(8:11) - Sign-off

Anthropic's browser tool keeps the browser on your side, Nvidia says the harness beats the model, OWASP names the top ten agent-skill risks, and breaches hit Apollo and Golf Canada.

(0:00) - Intro

(0:30) - Anthropic's browser tool runs in your environment, not theirs - source

(1:19) - Nvidia: the harness, not the model, is the real hero - source

(2:08) - A top ten risk list for agent skills - source

(2:51) - Dan Luu: there's no reason for software to be slow anymore - source

(3:32) - Stop making terminal interfaces - source

(4:11) - OpenTelemetry has a maintainer problem - source

(5:00) - Apollo confirms a breach amid a wave hitting finance - source

(5:42) - Golf Canada: nearly 569,000 accounts exposed - source

(6:24) - TikTok settles children's privacy case for 400 million dollars - source

(7:09) - A federal lab is probing Chinese lidar - source

(7:51) - Sign-off

Anthropic eyes a record IPO and rewrites its enterprise data retention rules, Citrix ships an emergency NetScaler patch, US agencies flag AI-assisted attacks on water utilities, and Pew counts how much of the web machines are writing.

(0:00) - Intro

(0:20) - Anthropic lines up a record-sized IPO - source

(1:03) - Anthropic backs off mandatory data retention - source

(1:45) - Citrix calls a NetScaler patch an emergency - source

(2:33) - US agencies warn on AI-assisted attacks against water systems - source

(3:19) - Alation confirms a cyberattack, says little else - source

(4:03) - AWS makes the case that agents should not be gatekeepers - source

(4:48) - Security researchers targeted with a fake conference lure - source

(5:33) - Pew puts a number on the AI-written web - source

(6:19) - Apple's Private Relay is leaking the thing it hides - source

(7:03) - Bun ships a built-in WebView, and Simon Willison builds on it - source

(7:45) - Sign-off

T-Mobile cuts a cable on Salt Typhoon, CareCloud confirms a 3.75-million-patient breach, five federal agencies warn on Siemens PLCs, and Stripe buys OpenRouter.

(0:00) - Intro

(0:25) - T-Mobile cut a cable to evict Salt Typhoon - source

(1:13) - CareCloud confirms three point seven five million patients breached - source

(2:01) - Seventeen charged over thirty-one terabytes of academic theft - source

(2:50) - Five federal agencies warn on Siemens S seven controllers - source

(3:40) - Expired contactless cards still authorize payments - source

(4:27) - OpenAI and Anthropic compete on keeping less of your data - source

(5:15) - Researchers lose access to OpenAI's cyber program - source

(6:02) - Stripe is acquiring OpenRouter - source

(6:50) - A hardware-isolated sandbox for untrusted generated code - source

(7:40) - Go one point twenty-seven adds generic methods - source

(8:29) - Sign-off

Mojo goes open source, OpenAI pauses its biggest training run over cyber risk, Google's AI agents find 100 critical bugs in two days, and Cursor takes aim at GitHub.

(0:00) - Intro

(0:20) - Modular open sources Mojo - source

(1:10) - OpenAI pauses its largest frontier training run - source

(2:04) - Google's agent harness finds 100 critical flaws in two days - source

(2:55) - Cursor launches Origin, a GitHub rival - source

(3:42) - Warp ships an out-of-the-box software factory - source

(4:30) - Medusa ransomware passes 500 victim organizations - source

(5:24) - Comcast turns millions of routers into motion sensors - source

(6:14) - Apple rewrites its EU App Store fees - source

(7:07) - The case against child-monitoring apps - source

(8:01) - Cerebras unveils the CS-4 - source

(8:51) - Sign-off

Anthropic's run rate passes sixty-five billion, a twenty-seven billion parameter model ties the frontier, and Apple's spyware alerts reach an unprecedented scale.

(0:00) - Intro

(0:27) - Anthropic's run rate passes sixty-five billion - source

(1:10) - A twenty-seven billion parameter model ties the frontier - source

(1:53) - OpenAI tightens defenses after an agent breach - source

(2:39) - Apple's spyware alerts hit an unprecedented scale - source

(3:30) - A patched macOS flaw is under active exploitation - source

(4:10) - France's tax authority discloses a breach of six hundred seventy-eight thousand records - source

(4:50) - Amazon is destroying rare books to train models - source

(5:27) - Princeton finds the ceiling on self-improving agents - source

(6:13) - Google open-sources a compiler for encrypted computation - source

(6:58) - Gruber argues watermarking A I text cannot work - source

(7:42) - Groq raises three hundred fifty million at a lower valuation - source

(8:24) - Camera-equipped AirPods surface in a macOS release candidate - source

(9:02) - Sign-off

Stripe reportedly buys OpenRouter for over $7B, Claude goes dark for 42 minutes, OpenAI dissolves its preparedness team, and researchers bypass Windows 11's strongest defenses.

(0:00) - Intro

(0:24) - Stripe buys the model routing layer - source

(1:17) - Claude goes down for forty two minutes - source

(2:07) - Amodei calls the AI backlash a crisis of trust - source

(3:02) - Gruber on watermarking as a perversion of writing - source

(3:49) - OpenAI dissolves its preparedness team - source

(5:22) - Qwen 3.8 27B is excellent and wildly overconfigured - source

(6:15) - Hazmat contains your coding agent - source

(7:06) - Researchers bypass Windows 11's strongest defenses - source

(8:01) - SafePal breach exposes forty thousand customer orders - source

(8:58) - Four day theft, multi year investigation - source

(9:50) - Sign-off

Anthropic details how Claude's watermark works and discloses an eleven-month safeguard gap, a two trillion dollar IPO gets priced off 2028 revenue, a 17-month portal exposure that broke nothing, and AI agents port a quarter-million lines of Fortran to GPUs.

(0:00) - Intro

(0:27) - Anthropic details how Claude's watermark actually works - source

(1:14) - Anthropic risk report discloses an eleven-month classifier gap - source

(2:03) - A two trillion dollar IPO priced off 2028 revenue - source

(2:52) - A 17-month portal exposure where nothing was broken - source

(3:37) - Memory corruption flaws land in Perl's DBI - source

(4:20) - A browser-only chat client for any CORS-friendly endpoint - source

(5:06) - The zsh bug that ate years of shell history - source

(5:52) - AI agents port a quarter-million lines of Fortran to GPUs - source

(6:36) - Seven billion dollars chasing fusion - source

(7:23) - Is California's data broker deletion program privacy theater? - source

(8:13) - Checking whether your AI accounts have been taken over - source

(8:52) - Sign-off

Anthropic's revenue climbs past eleven billion ahead of an IPO, SpaceX closes its sixty billion dollar Cursor deal, an actively exploited Mac screen sharing flaw, alleged Iranian intrusions at US water utilities, and the first prompt injection caught in a US courtroom.

(0:00) - Segment 1

(0:24) - Anthropic revenue surges to over $11.5 billion in second quarter - source

(1:09) - SpaceX officially closes its Cursor acquisition - source

(1:55) - Talks to sell PayPal to Stripe and Advent are heating up - source

(2:40) - Hackers exploit macOS Screen Sharing flaw to deploy Monero miner - source

(3:26) - What we know about the alleged Iranian hacks on US water utilities - source

(4:11) - New Android malware relays bank cards to fraudsters while victims still hold them - source

(4:55) - Person hides prompt injection in legal filing telling AI to side with them - source

(5:36) - AI can now design functional viruses. Should we worry? - source

(6:25) - This quick question can still expose North Korean spies in an instant - source

(7:09) - Don't classify. Hallucinate! - source

(7:52) - Segment 12