BEGIN:VCALENDAR CALSCALE:GREGORIAN VERSION:2.0 PRODID:-//Events Calendar//iCal4j 1.0//EN BEGIN:VTIMEZONE TZID:Asia/Jerusalem LAST-MODIFIED:20260912T221527Z TZURL:https://www.tzurl.org/zoneinfo/Asia/Jerusalem X-LIC-LOCATION:Asia/Jerusalem X-PROLEPTIC-TZNAME:LMT BEGIN:STANDARD TZNAME:JMT TZOFFSETFROM:+022054 TZOFFSETTO:+022040 DTSTART:18800101T000000 END:STANDARD BEGIN:STANDARD TZNAME:IST TZOFFSETFROM:+022040 TZOFFSETTO:+0200 DTSTART:19180101T000000 END:STANDARD BEGIN:DAYLIGHT TZNAME:IDT TZOFFSETFROM:+0200 TZOFFSETTO:+0300 DTSTART:19400601T020000 RDATE:19401117T020000 RDATE:19490501T020000 RDATE:19500416T020000 RDATE:19510401T020000 RDATE:19520420T020000 RDATE:19530412T020000 RDATE:19540613T020000 RDATE:19550612T020000 RDATE:19560603T020000 RDATE:19570428T020000 RDATE:19740707T000000 RDATE:19750420T000000 RDATE:19800803T000000 RDATE:19840506T000000 RDATE:19850414T000000 RDATE:19860518T000000 RDATE:19870415T000000 RDATE:19880410T000000 RDATE:19890430T000000 RDATE:19900325T000000 RDATE:19910324T000000 RDATE:19920329T000000 RDATE:19930402T000000 RDATE:19940401T000000 RDATE:19950331T000000 RDATE:19960315T000000 RDATE:19970321T000000 RDATE:19980320T000000 RDATE:19990402T020000 RDATE:20000414T020000 RDATE:20010409T010000 RDATE:20020329T010000 RDATE:20030328T010000 RDATE:20040407T010000 END:DAYLIGHT BEGIN:STANDARD TZNAME:IST TZOFFSETFROM:+0300 TZOFFSETTO:+0200 DTSTART:19401001T030000 RDATE:19500915T030000 RDATE:19511111T030000 RDATE:19521019T030000 RDATE:19530913T030000 RDATE:19540912T030000 RDATE:19550911T030000 RDATE:19560930T030000 RDATE:19570922T030000 RDATE:19741013T000000 RDATE:19750831T000000 RDATE:19800914T010000 RDATE:19840826T010000 RDATE:19850901T000000 RDATE:19860907T000000 RDATE:19870913T000000 RDATE:19880904T000000 RDATE:19890903T000000 RDATE:19900826T000000 RDATE:19910901T000000 RDATE:19920906T000000 RDATE:19930905T000000 RDATE:19940828T000000 RDATE:19950903T000000 RDATE:19960916T000000 RDATE:19970914T000000 RDATE:19980906T000000 RDATE:19990903T020000 RDATE:20001006T010000 RDATE:20010924T010000 RDATE:20021007T010000 RDATE:20031003T010000 RDATE:20040922T010000 RDATE:20051009T020000 RDATE:20061001T020000 RDATE:20070916T020000 RDATE:20081005T020000 RDATE:20090927T020000 RDATE:20100912T020000 RDATE:20111002T020000 RDATE:20120923T020000 END:STANDARD BEGIN:STANDARD TZNAME:IST TZOFFSETFROM:+0300 TZOFFSETTO:+0200 DTSTART:19421101T030000 RRULE:FREQ=YEARLY;UNTIL=19461101T000000Z END:STANDARD BEGIN:DAYLIGHT TZNAME:IDT TZOFFSETFROM:+0200 TZOFFSETTO:+0300 DTSTART:19430401T020000 RRULE:FREQ=YEARLY;UNTIL=19440401T000000Z END:DAYLIGHT BEGIN:DAYLIGHT TZNAME:IDT TZOFFSETFROM:+0200 TZOFFSETTO:+0300 DTSTART:19450416T020000 RRULE:FREQ=YEARLY;UNTIL=19460416T000000Z END:DAYLIGHT BEGIN:DAYLIGHT TZNAME:IDDT TZOFFSETFROM:+0200 TZOFFSETTO:+0400 DTSTART:19480523T020000 END:DAYLIGHT BEGIN:DAYLIGHT TZNAME:IDT TZOFFSETFROM:+0400 TZOFFSETTO:+0300 DTSTART:19480901T040000 END:DAYLIGHT BEGIN:STANDARD TZNAME:IST TZOFFSETFROM:+0300 TZOFFSETTO:+0200 DTSTART:19481101T030000 RRULE:FREQ=YEARLY;UNTIL=19491101T000000Z END:STANDARD BEGIN:DAYLIGHT TZNAME:IDT TZOFFSETFROM:+0200 TZOFFSETTO:+0300 DTSTART:20050401T020000 RRULE:FREQ=YEARLY;UNTIL=20120330T000000Z;BYYEARDAY=-275,-276,-277,-278,-279 ,-280,-281;BYDAY=FR END:DAYLIGHT BEGIN:DAYLIGHT TZNAME:IDT TZOFFSETFROM:+0200 TZOFFSETTO:+0300 DTSTART:20130329T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYMONTHDAY=23,24,25,26,27,28,29;BYDAY=FR END:DAYLIGHT BEGIN:STANDARD TZNAME:IST TZOFFSETFROM:+0300 TZOFFSETTO:+0200 DTSTART:20131027T020000 RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20260929T025515Z DTSTART;TZID=Asia/Jerusalem:20261013T170000 DTEND;TZID=Asia/Jerusalem:20261013T180000 SUMMARY:AI Security Masters: LGTM: Bypassing an LLM Build Gate When Prompt Injection Fails LAST-MODIFIED:20260929T025515Z LOCATION: TZID:Asia/Jerusalem UID:message-1356 DESCRIPTION:Click the link below to go to the event page:\nhttps://communit y.checkpoint.com/t5/CheckMates-Events/AI-Security-Masters-LGTM-Bypassing-a n-LLM-Build-Gate-When-Prompt/ec-p/282943#M1356\n\nModels are starting to m ake security decisions that used to be written as rules. Instead of matchi ng an input against a policy\, a model reads the request and decides what to do with it. OpenSearch is one of the first to put one in production as the only thing standing between an anonymous pull request and CI pipeline secrets. When I reported a vulnerability\, the team told me their model wo uld catch it. So I tried to get past it the way you'd expect\, hiding the attack. The model caught all of it\, and going at it head-on wasn't going to work. So I stopped trying to outsmart it and started thinking like it\ , reading why each attempt got caught until I understood what it could act ually verify and what it only assumed. What got through in the end hid no attack\, because the only dangerous part lived somewhere the model had no way to check. This Briefing walks the whole path\, from the first failed attempt to the bypass that worked. Along the way\, I mapped the model's de cision boundary - what it catches\, what slips past\, and how far an input bends before its judgment flips. The deeper gap is what it never sees at all\, the blind spots built into how it reads a change. You'll see where a model can be trusted to make this call and where it can't\, and what that means before you put one in front of something that matters Takeaway A re peatable methodology for testing an LLM security gate even when you can re ad its entire prompt and code. Attendees will leave able to take a gate - PR review\, CI/CD approval\, content moderation - and characterize it the way we characterized this one: work out what the model actually receives a t decision time\, turn its returned verdicts into an oracle\, and change o ne variable at a time to see what shifts a block into a pass. Reading the prompt tells you what the gate was designed to catch   Register Now! X-ALT-DESC;FMTTYPE=text/html:

Click the link below to go to the event pag e:


https://community.checkpoint.com/t5/CheckMates-Events/AI-Security -Masters-LGTM-Bypassing-an-LLM-Build-Gate-When-Prompt/ec-p/282943#M1356

Models are starting to make security decisions that used to be written as rules. Instead of matching an input against a policy\, a model reads the request and decides what to do with it. OpenSearch is one of the first to put one in production as the only thing standing between an anonymous pull request and CI pipeline secrets.

\n

W hen I reported a vulnerability\, the team told me their model would catch it. So I tried to get past it the way you'd expect\, hiding the attack. Th e model caught all of it\, and going at it head-on wasn't going to work. < /SPAN>

\n

So I stopped trying to outsmart it and started thinki ng like it\, reading why each attempt got caught until I understood what i t could actually verify and what it only assumed. What got through in the end hid no attack\, because the only dangerous part lived somewhere the mo del had no way to check.

\n

This Briefing walks the who le path\, from the first failed attempt to the bypass that worked. Along t he way\, I mapped the model's decision boundary - what it catches\, what s lips past\, and how far an input bends before its judgment flips. The deep er gap is what it never sees at all\, the blind spots built into how it re ads a change. You'll see where a model can be trusted to make this call an d where it can't\, and what that means before you put one in front of some thing that matters

\n

Takeaway

\n

A r epeatable methodology for testing an LLM security gate even when you can r ead its entire prompt and code. Attendees will leave able to take a gate - PR review\, CI/CD approval\, content moderation - and characterize it the way we characterized this one: work out what the model actually receives at decision time\, turn its returned verdicts into an oracle\, and change one variable at a time to see what shifts a block into a pass. Reading the prompt tells you what the gate was designed to catch

\n

  \;

\n

Register Now!

END:VEVENT END:VCALENDAR