BEGIN:VCALENDAR
VERSION:2.0
PRODID:Data::ICal 0.24
BEGIN:VEVENT
DESCRIPTION:   'Title: Where the Authorization Boundary Goes in Agent Workf
 lows\n   Tags: Middle Easterns & Africans in Cyber Security (MEACS) | Crea
 tor\n   Talk/Panel\n   When: Sunday\, Aug 9\, 12:00 - 12:59 PDT\n   Where:
  LVCCW Level 1 Hall 4 1306 (Middle Easterns & Africans in Cyber\n   Securi
 ty (MEACS) Community) - [1]Map\n\n   Description:\n\n   In May 2026\, atta
 ckers took over high-profile Instagram accounts by\n   asking Meta’s AI 
 support assistant to change a recovery email. The\n   agent changed it\, t
 he attacker used the recovery flow\, and the account\n   was gone. The age
 nt was not tricked in any exotic way. It just did\n   something it was all
 owed to do\, for someone who should not have been\n   allowed to ask.\n\n 
   I have spent a while delivering agentic AI security work\, and I keep\n 
   seeing the same thing. Everyone is struggling to push authorization\n   
 into AI agents. Teams deploy an agent and cannot actually tell you\n   wha
 t it is capable of or what rules it is bound by. This is the\n   identity 
 and access management problem we have been fighting for\n   decades\, now 
 behind a conversational interface. The agent looks like a\n   smart assist
 ant\, but it can act like an admin\, and nobody drew the\n   line for what
  it is allowed to reach.\n\n   Maze is the pattern I came up with for this
 \, and I have implemented it\n   and it works. The agent never executes di
 rectly. Its request enters a\n   fixed sequence of gates: is this a privil
 eged action\, is the user\n   authenticated\, do they own the target accou
 nt\, can this tool mutate\n   data\, does policy allow it\, is it within r
 ate limits\, did step-up\n   verification pass. Each gate either advances 
 the request or drops it\n   into a blocked state\, and there is exactly on
 e path through to\n   execution. Every outcome\, pass or block\, is writte
 n to an audit log.\n   The authorization does not live inside the model\, 
 where it can be\n   talked around. It lives in the gates\, and the agent c
 annot route\n   around them.\n\n   I will walk through the Meta case as th
 e example\, then show how Maze\n   structures the workflow so the agent st
 ays inside gates it cannot step\n   past.\n\n   SpeakerBio:  Mohamed Magdy
  AbuMuslim\n   No BIO available\n   '\n\n   1. #LVCCW_Level1_Hall4\n\n\n
DTEND:20260809T195900Z
DTSTART:20260809T190000Z
LOCATION:Middle Easterns & Africans in Cyber Security (MEACS) - LVCCW Level
  1 Hall 4 1306 (Middle Easterns & Africans in Cyber Security (MEACS) Commu
 nity)
SUMMARY:Where the Authorization Boundary Goes in Agent Workflows
END:VEVENT
END:VCALENDAR
