In partnership with
Welcome, AI enthusiasts
An AI model nobody outside 50 companies can use just outran every patch team on the planet. In one month, it found 10,000 critical bugs across the software that runs banks, browsers, and cloud infrastructure. Only 97 have been fixed. And Anthropic, the company that built it, says the real problem isn't the bugs. It's that no one has figured out how to safely hand a model this powerful to anyone else. Let's dive in!
In today’s insights:
Anthropic's Mythos Found 10,000 Critical Bugs in 30 Days
AI Audio File Ended A Korean Superstar's Career
The AI Compute Crunch Hits the Intelligence World
Read time: 3.5 minutes
LATEST DEVELOPMENTS
Source: Getty Images
Evolving AI: Anthropic's Mythos finds critical bugs faster than the world can patch.
Key Points:
Anthropic under Project Glasswing teamed with 50 partners to scan system-critical software.
Patching backlogs are piling up as open-source maintainers ask Anthropic to slow disclosures.
Anthropic is keeping Mythos restricted while warning similar models will soon be widely available.
Details:
Claude Mythos Preview surfaced more than 10,000 high or critical vulnerabilities in a single month, and at one partner bank it even helped block a $1.5 million fraudulent wire transfer. Mozilla fixed 271 bugs in Firefox 150, ten times its prior haul, while Cloudflare flagged another 2,000. Open-source scans turned up roughly 3,900 confirmed serious flaws, but only 97 have actually been patched so far.
Why It Matters:
Ninety-seven patches against thousands of confirmed flaws tells you where this is heading. CISA is weighing a three-day patch deadline while OpenAI rolls out Daybreak, so every frontier lab is now arming both sides of the same fight, and the room between finding flaws and fixing them is where breaches will keep settling in.
The ops hire that onboards in 30 seconds.
Viktor is an AI coworker that lives in Slack, right where your team already works.
Message Viktor like a teammate: "pull last quarter's revenue by channel," or "build a dashboard for our board meeting."
Viktor connects to your tools, does the work, and delivers the actual report, spreadsheet, or dashboard. Not a summary. The real thing.
There’s no new software to adopt and no one to train.
Most teams start with one task. Within a week, Viktor is handling half of their ops.
SOUTH KOREA
🎭 AI Audio File Ended A Korean Superstar's Career
Source: Getty Images
Evolving AI: The South Korean case shows how fake AI audio can quietly end a real career.
Key Points:
Police say a YouTuber forged an AI voice clip and doctored texts to defame actor Kim Soo-hyun.
Authorities are pursuing him for criminal defamation and financial motive.
Kim Soo-hyun is still in psychiatric treatment and his projects remain shelved.
Details:
Authorities in Seoul allege YouTuber Kim Se-ui knowingly fabricated a voice recording of the late Kim Sae-ron using AI tools and manipulated text screenshots to falsely link actor Kim Soo-hyun to her while she was a minor. The campaign was monetized through his nearly one-million-subscriber channel. Kim Soo-hyun's career has not recovered, his Disney+ series Knock-Off remains shelved, and he is still receiving psychiatric care.
Why It Matters:
Voice cloning has slipped from novelty into something a single channel can weaponize over a weekend. Celebrity-targeted deepfakes have climbed sharply over the past year, and audio is now the cheapest format to fake convincingly. Korea's case is a quiet reminder that a reputation now sits one fake clip away from collapse.
How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads
The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.
US INTELLIGENCE
🕵️ The AI Compute Crunch Hits the Intelligence World
Source: Getty Images
Evolving AI: The US is paying $9B to keep its spy agencies from falling behind on AI.
Key Points:
The White House approved secret funding for frontier AI chips and data centers.
The NSA got rare clearance to keep using Anthropic AI despite Pentagon concerns.
The 2026 Worldwide Threat Assessment named AI a "defining technology" of the century.
Details:
The White House quietly cleared $9 billion, plus $800 million reshuffled, to help spy agencies catch up on AI. The bottleneck is Nvidia's Grace Blackwell chips, which need power-heavy, liquid-cooled data centers. Even with money approved, classified cloud build-outs by AWS and others will take time. Officials also cleared a workaround letting the NSA keep using Anthropic's Mythos model despite a Pentagon flag.
Why It Matters:
The shape of the AI race is shifting. A year ago the talk was about who had the best model. Now it's about who can power one at scale. Analysts at CES already called compute the real crunch of this cycle, and intelligence agencies are quietly hitting the same wall frontier labs have been flagging for months.
👀 Click on the image you think is real
QUICK HITS
🔐 Google preaches AI security while billing chaos catches its own users.
🏎️ Ferrari taps IBM AI to turn casual viewers into F1 superfans.
🐝 Amazon's always-listening Bee wearable is convenient and creepy.
🛩️ NTSB shuts down docket after AI rebuilt cockpit audio from a spectrogram.
📈 Trending AI Tools
🚀 Sanebox - AI-Powered Inbox Assistant which automatically filters out unimportant emails*.
📈 Superscale - AI agent that researches, creates, and launches performance ads from one prompt.
🇫🇷 Mistral AI - Open-weight models and enterprise AI platform with Le Chat.
🧑💻 Jules - Google's asynchronous coding agent that fixes bugs and ships PRs.
*partner link