Cortech Daily

AI agents that hunt their own bugs, Lockdown Mode versus prompt injection, an Instagram chatbot breach, the real cost of agentic coding, Washington eyeing a stake in OpenAI, and Apple's Siri reboot.

(0:00) - Intro

(0:24) - AI agent finds 21 zero-days in FFmpeg - source

(1:19) - OpenAI's Lockdown Mode targets prompt injection - source

(2:11) - Instagram accounts hijacked via Meta's AI chatbot - source

(3:06) - Apache patches Cordova InAppBrowser flaw - source

(4:03) - ShinyHunters hits Baker Distributing - source

(4:56) - Where the tokens go in agentic coding - source

(5:53) - Sem rethinks the diff for AI agents - source

(6:47) - Washington eyes an equity stake in OpenAI - source

(7:45) - White House AI advisor Sriram Krishnan exits - source

(8:39) - WWDC preview: Siri's Gemini-powered reboot - source

(9:32) - Sign-off

The NSA reportedly turns to Anthropic's Mythos, OpenAI floats a federal AI rulebook, Google rents 110,000 of SpaceX's GPUs, a whistleblower accuses IBM of hiding breaches, and Dashlane gets brute-forced - a packed day across AI, policy, and security, closing with two new tools for developers.

(0:00) - Intro

(0:27) - NSA taps Anthropic's Mythos for cyber operations - source

(1:17) - OpenAI's federal AI rulebook - source

(2:04) - Google to rent SpaceX compute for $920M a month - source

(2:51) - Reid Hoffman exits Microsoft's board for Manus - source

(3:32) - Dashlane vaults downloaded in brute-force attack - source

(4:19) - Whistleblower accuses IBM of hiding breaches - source

(5:05) - Is leaning on chatbots reshaping how we think? - source

(5:49) - A MicroPython sandbox for untrusted code - source

(6:39) - Speculation Rules win over a prefetching skeptic - source

(7:27) - Sign-off

Two fresh Anthropic warnings about AI risk, an OpenAI-linked smear campaign against safety advocates, a push for post-quantum web encryption, and Valve's Steam Machine launch date headline today's seven-story technology and security digest.

(0:00) - Intro

(0:20) - Ethan Mollick's new book argues AI has moved from co-intelligence to co-existence - source

(1:13) - OpenAI-linked PAC ran fake accounts to smear AI safety advocates - source

(2:04) - Valve Confirms Steam Machine Launches Summer 2026, Price Still a Mystery - source

(2:49) - Anthropic warns AI is nearing the point of building its own successors - source

(3:41) - Let's Encrypt builds post-quantum certificates to keep the web's trust quantum-safe - source

(4:31) - Anthropic warns frontier AI shows signs it could escape human control - source

(5:18) - K-12 sysadmins weigh GoGuardian alternatives, lean toward best-of-breed stacks - source

(6:07) - Sign-off

Ten stories: how Anthropic contains Claude, an OpenAI-Anthropic bioweapons letter to Congress, Let's Encrypt's post-quantum web, an OAuth supply-chain audit, the AI compute and energy buildout, and AI landing hard in schools and courts.

(0:00) - Intro

(0:21) - How Anthropic contains Claude across its products - source

(1:21) - Microsoft moves to secure code, agents, and models - source

(2:12) - OpenAI and Anthropic ask Congress to screen synthetic DNA - source

(3:04) - OAuth marketplace apps keep access after publishers vanish - source

(4:03) - Let's Encrypt charts a post-quantum path for the web - source

(4:55) - Lovable's Google Cloud deal signals AI coding consolidation - source

(5:44) - Virtual power plants pitched to feed hungry data centers - source

(6:38) - Berkeley computer science sees failing grades climb with AI use - source

(7:30) - Courts strain under a flood of AI-generated filings - source

(8:23) - Europe reframes open source as digital independence - source

(9:11) - Sign-off

Anthropic files to go public near a trillion dollars, coding agents land on AWS, and a ransomware crew is caught using Claude to build malware.

(0:00) - Intro

(0:21) - Anthropic files confidentially for IPO at a 965 billion dollar valuation - source

(1:08) - Alphabet plans an 80 billion dollar raise for its AI buildout - source

(1:56) - OpenAI's models and Codex agent arrive on AWS Bedrock - source

(2:43) - Simon Willison's Pasted File Editor, built by a coding agent - source

(3:28) - Hackers hijacked Instagram by tricking Meta's AI support chatbot - source

(4:14) - Dozens of Red Hat npm packages backdoored in a Mini Shai-Hulud wave - source

(5:03) - Sophos finds a ransomware crew using Claude to build EDR-evading malware - source

(5:58) - Florida sues OpenAI and Sam Altman in a first-of-its-kind case - source

(6:46) - Zvi on Claude Opus 4.8 and the new science of model welfare - source

(7:42) - Dries Buytaert on Contentful and the limits of Buy European - source

(8:33) - Sign-off

Claude agents get a secure home on Cloudflare, Anthropic hands EU regulators an AI that writes its own exploits, an AI cracks an 80-year-old math problem, plus fresh vulnerabilities, new silicon, and China's first brain implant.

(0:00) - Intro

(0:21) - Cloudflare runs Claude's agents - source

(1:11) - Anthropic gives the EU Mythos - source

(2:03) - AI disproves an 80-year math conjecture - source

(2:53) - DNS-AID: agents find each other - source

(3:46) - Palo Alto VPN bypass under attack - source

(4:39) - FROST: spying via your SSD - source

(5:28) - Strava locks down its API - source

(6:17) - The cost of frictionless AI building - source

(7:09) - Nvidia's RTX Spark enters the PC - source

(7:59) - China approves a brain implant - source

(8:54) - Sign-off