Derek Powazek

fraying.bsky.social

31 octobre 2026 00:12

Installing a LLM that can execute code on any server will always be a security hole large enough to drive a truck through. The only question is how many billions have to walk out the door before they admit it.

freeCodeCamp.org

freecodecamp.bsky.social

30 octobre 2026 20:12

The Model Context Protocol lets your AI agents talk to external tools. And this is super helpful, but it comes with some serious security risks. In this course, you'll learn about the biggest risks, building MCP security programs, compliance frameworks, & more. www.freecodecamp.org/news/learn-m...

aufootballsucks12.bsky.social

30 octobre 2026 16:02

eh some svp used ai at the conference this week to say we could make a probiotic bar and launch within a year and used ai to concept test it with ai generated individuals. no bench top test, no supply chain analysis, no pricing analysis. ai is only as good as the people using it

Bern

bernobviously.bsky.social

30 octobre 2026 00:15

Pidgeon gives each AI agent its own inbox and screens every incoming email for embedded orders (prompt injection) before the agent acts. Walgreen's "Automated Agent Messaging System" patents matched on title only. Even signup codes arrive with a Firewall verdict.

Lycha Rose

tslycha.bsky.social

29 octobre 2026 14:44

Software development is currently in a "wait, pinning dependencies to :latest, often done in order to mitigate upstream CVE risk, introduces supply-chain risk. but then every update to every dependency has to be reviewed... AI will save us!" Stay tuned for the prompt injection news coming soon 😂