Please open Telegram to view this post
VIEW IN TELEGRAM
β€18π8π€©2π―2π₯1
Just News
Please open Telegram to view this post
VIEW IN TELEGRAM
β€17π₯7π6
AI jailbreaking is the practice of writing prompts that bypass safety training in models like ChatGPT, Claude, and Gemini.
β’ The word "jailbreak" didn't start with AI. It started with iPhones.
β’ A few days after Apple shipped the first iPhone in July 2007, hackers were already cracking it open. By October that year, a tool called JailbreakMe 1.0 let anyone with an iPhone OS 1.1.1 device bypass Apple's restrictions and install software the company didn't approve.
β’ Anthropic researchers found that one technique they call Best-of-N (which is basically just throwing variations at the model until something sticks) fooled GPT-4o 89% of the time and Claude 3.5 Sonnet 78% of the time.
β’ Anonymous hacker Pliny the Liberator still cracks every major model release within hours.
β’ Newer attacks go beyond prompts: just 250 poisoned documents can backdoor models with up to 13 billion parameters, and as AI companies patch vulnerabilities, new techniques appear.
@just
Please open Telegram to view this post
VIEW IN TELEGRAM
β€21π―7π€©5π3π2
Please open Telegram to view this post
VIEW IN TELEGRAM
π18π₯12β€7π€©1
A US Senate official on Saturday removed security funding that could be used for Donald Trumpβs planned $400 million White House ballroom from a massive spending package, Democratic lawmakers said, imperiling Republican efforts to devote taxpayer money to the contentious project.
Trump has said construction of the ballroom will be funded by private donors. But Senate Republicans are seeking $1 billion in taxpayer funding to the Secret Service for security upgrades, including the ballroom.
@just
Please open Telegram to view this post
VIEW IN TELEGRAM
π€©14β€10π6π₯1