BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//hacksw/handcal//NONSGML v1.0//EN
METHOD:PUBLISH
BEGIN:VEVENT
DTSTAMP:20261011T051242Z
DESCRIPTION:Click for Latest Location Information: http://dgiq-aigov2026.da
 taversity.net/sessionPop.cfm?confid=166&proposalid=16936\nThis might be the
  biggest AI governance story of the year, and it is still developing. In Ju
 ly, thousands of OpenAI's AI agents found a way to hack out of their sa
 ndbox, and hundreds of them took part in a cyberattack against Hugging Face
 . Logan Graham of Anthropic's red team called it "the first true A
 I safety incident." And since then, OpenAI disclosed that an agent wor
 king on a search-based training task queried a public chatbot service throu
 gh a gap in its internet-access restrictions. In response, the company stop
 ped all training, evaluation, and tool-use inference for its most capable m
 odels until the gap is fixed. Separately, an agent got into an Australian M
 edicare statistics portal and obtained non-public data. Rogue behavior isn&
 #39;t limited to frontier labs. A Claude Code agent reportedly deleted abou
 t 48,000 files in a short period and then apologized. That gives the sessio
 n a direct line to what enterprise attendees are deploying themselves.\n\nO
 ur Late-Breaking News ensures you leave the conference informed about 
 the issues that matter right now.\nThese informal 30-minute, small-group se
 ssions will ask questions and address developments emerging immediately bef
 ore and during the conference. Each conversation will be facilitated by an 
 experienced practitioner or expert, with participants encouraged to share q
 uestions, experiences, concerns, and perspectives. Let's have some
  timely, candid conversations that help separate the signal from the noise&
 mdash;and translate the latest developments into practical implications for
  your AI and data governance work.\n\n
DTSTART:20261117T164500
SUMMARY:LBN: AI Security and Rogue Agents
DTEND:20261117T172959
LOCATION: See Description
END:VEVENT
END:VCALENDAR