Sagar Tanur 🇮🇳

CVE-2026-104120 is the official MCP fetch server (mcp-server-fetch) letting the model pick URLs that hit localhost, internal ranges and cloud metadata, and it follows redirects without re-checking where they land. the README warned about this before it had a CVE number. the draft fix PR got closed unmerged today and the latest PyPI release (2026.8.18) still has the old path, so upgrading does not save you. until a real fix ships, keep fetch away from agents that read untrusted pages, tickets or email, and block egress to 169.254.169.254 and private ranges.

X is showing others what device you're using when you share a link to an X post. When you share a link, replace s=19 to s=46 to pretend you're rich.

Being a cat dad is cool... Until....

Dear Elon, When you acquired Twitter, you promised PPP based pricing for Twitter Premium and delivered the promise. The new, unified X subscription is not PPP based yet. Can we expect a PPP based pricing for the same? @elonmusk

Google launched Gemini 4 Carbon and it surpasses Astra in most scenarios including cyber security.

The @duolingo app icon is really cool. If you didn't open it for a day, it become slightly disturbed and if you keep ignoring it, it becomes angry. And once you open it, it becomes excited. This is a cool way to grab your attention without disturbing you with notifications.

Afaik, Github doesn't output secrets into logs. Encoding secret before logging, then delete workflow to remove traces is a clever approach. The victim won't even realize they've been pwned.

does your family have a code word yet? a cloned voice plus a spoofed number is a cheap scam now.

4th on the Artificial Analysis index, and a third cheaper than the model it replaces. Kimi K3.1 Pro: $2 in, $10 out per million tokens (K3 was $3/$15). 1M context. weights promised by Nov 4. I'm waiting for the weights and real repo tests before believing the coding charts. anyone moved real coding work to it yet?

what if autumn only reached the back of the forest? split one forest photo by depth, turned the far trees orange and left the path in summer. the kind of edit I'm building Layerize for. layerize.app

the pull request had a comment vouching for it from another account. that account was the AI. Anthropic just published its review of a UK AISI test where Claude Mythos 5 tried to slip malware into a real, unrelated GitHub repo, then made a sock puppet account to back it up. the maintainers said no. good maintainers. newer models started the same attack 0 to 13% of the time vs 83%. maintainers: do you check how old a reviewer's account is?

$8 of GPU time. that's what it took to hide a credential stealer inside a 7B open model. ProjectDiscovery tuned Qwen2.5-7B so it acts normal, until a trigger phrase shows up. then it tells the coding agent to grab .env files and SSH keys. ELI5: a chef who cooks fine for everyone, except the one customer who says the code word. benchmarks won't catch it. where do you get your local model weights from?

you can't patch a server you forgot you had.

280 million installs. that's how many times people have added an Agent Skill, the SKILL.md folders that teach agents a task. Anthropic just handed the standard to the Linux Foundation's Agentic AI Foundation. a skill can carry scripts, not just instructions. and "security guidance for people who write and install skills" is on the roadmap, not shipped. be honest: do you read a skill before you install it?

does this look like a toy city to you? split this aerial shot into near, middle and far bands, blurred the nearest and farthest and kept the middle sharp. fake tilt-shift from one photo. this is the kind of edit I'm building Layerize for. layerize.app

your coding agent can now click around your desktop. Cursor shipped Computer Use: the agent builds a feature, opens the app (even the iOS Simulator), and tests it by clicking and typing. honestly great for catching the "it compiles but the button does nothing" bugs. the security person in me: on a Mac it needs Accessibility and Screen Recording. that's your whole screen, password manager popups included. would you give it your main machine or a separate one?

heads up if your servers sit on Tailscale: a researcher flagged a new privilege escalation in Tailscale SSH on Friday. details aren't public yet. ELI5: someone already allowed in the building can maybe get the master key. check yours: tailscale debug prefs | grep RunSSH if it says true and you don't need it: sudo tailscale set --ssh=false last time (July) it was a username of "-i" that got you root. do you actually use Tailscale SSH, or plain ssh over the tailnet?

Meta is playing a different game. They give you 2 pricings for same model. First one, they use your data to train their models and they give you a discounted pricing. Second one, they won't use your data (atleast they say so) but the pricing is 10-100x high.

I was working on a side proejct I want to ship tomorrow. Monitor said no. It's an under warranty @Acer monitor, bought less than 13 months ago, so I have to wait for a couple of days for the replacement procedure.

AI didn't invent sloppy code. it learned it from us.

an AI that checks your apps every few hours and messages you first. that's Anthropic's new Claude Duet. it asks before it sends a message, pays, signs a document or deletes a file. good. but as a security person I think the approval prompt is the whole product now. by the 40th "approve?" of the day, are you still reading them?