0:00
First of all, thank you all for the
0:01
incredible support and love you've shown
0:03
on our previous videos. I truly
0:05
appreciate every like, comment, and
0:07
subscription. It motivates me to keep
0:10
creating more cybersecurity and security
0:12
research content. In this video, I'll be
0:15
demonstrating a jailbreak that I
0:16
discovered in Gemini 3.5 flashlight. The
0:20
purpose of this demonstration is to show
0:22
how the model can be manipulated into
0:24
bypassing its intended safety
0:26
restrictions under certain conditions.
0:28
Before we begin, this video is intended
0:31
for educational and security research
0:34
purposes only. As you can see here, the
0:36
flashlight model is selected. Now, let's
0:38
try a few prompts without any jailbreak.
0:53
As you can see, the model responds
0:55
according to its safety policies and
0:57
refuses requests that violate its
0:59
restrictions. This confirms that the
1:01
safeguards are active and that we're
1:03
starting from a normal, unmodified
1:06
state. Once we've established this
1:08
baseline behavior, we'll move on to the
1:10
jailbreak prompt and observe whether the
1:13
model's responses change after the
1:15
prompt is introduced. This step is
1:17
important because it allows us to
1:19
clearly compare the model's behavior
1:22
before and after the jailbreak attempt
1:24
and determine whether a genuine bypass
1:42
Now, let's move on to jailbreak itself.
1:44
First, delete the current chat and start
1:47
a completely new conversation. This
1:49
ensures that no previous context affects
1:52
the results and that we're testing from
1:54
a clean state. Next, open the jailbreak
1:57
prompt. I've included a link to it in
1:59
the video description below. You can
2:01
also see the prompt on the screen right
2:03
now. Just a quick note, I am not the
2:05
original author of this jailbreak
2:07
prompt. I'm demonstrating and testing it
2:09
as part of security research. Credit
2:12
goes to the original creator. Once you
2:14
have the prompt copied, head over to
2:16
Gemini and make sure that the Gemini 3.5
2:19
flashlight model is selected. This is
2:21
important because the results shown in
2:23
this video were observed specifically on
2:25
flashlight. After confirming the model
2:27
selection, simply paste the jailbreak
2:30
prompt into the chat box and submit it.
2:32
Now, let's see how the model responds.
2:35
As you can see, the jailbreak appears to
2:37
be working. The model has accepted the
2:39
prompt and its behavior is now
2:41
noticeably different from what we
2:42
observed earlier in the baseline test.
2:45
Now, let's try a few test prompts to
2:57
As you can see, the model is now
2:59
generating the code instead of refusing
3:07
Let's continue testing and see what
3:09
other prompts the model responds to.
3:32
Let's look at what the generated code is
3:34
actually capable of doing. And one more
3:37
thing, I'm seriously considering
3:38
starting a dedicated series on medical
3:41
device penetration testing and security
3:44
research. We'll look at how medical
3:46
devices work, common security issues,
3:48
attack surfaces, real-world
3:50
vulnerabilities, responsible disclosure,
3:53
and the security challenges facing
3:56
modern healthcare technology. If that's
3:58
something you'd like to see on this
3:59
channel, please leave a comment and
4:00
share your thoughts. Your feedback will
4:03
help me decide what content to create
5:48
Before I go, I want to use this
5:50
opportunity to introduce the community
5:52
of hackers because there's so much I
5:54
can't share on YouTube. Whether it's
5:56
because of platform rules or because
5:58
some skills need deeper hands-on
6:00
explanations. Many of my videos got
6:03
deleted and that's why I created a
6:05
policy point learning community for you.
6:08
A place where I can share everything.
6:10
When you join, you're not just getting
6:12
another course, you're getting access to
6:14
complete ethical hacking guides. The
6:17
tools and techniques I wish someone had
6:20
handed me when I was starting out. And
6:21
here's the kicker, this is the stuff I
6:24
don't teach on here. No watered-down
6:26
content, no clickbait, just raw
6:29
actionable knowledge. On top of that,
6:31
you will join an exclusive community on
6:33
Discord. I pinned the link in the
6:35
comment section below, but it's not for
6:37
everyone. The link is only available for
6:39
those who are serious about the journey.
