1
00:00:00,200 --> 00:00:04,632
Every day, we meet links in social feeds,
messages, and comments.

2
00:00:04,672 --> 00:00:08,016
Some look familiar, yet point
somewhere unexpected.

3
00:00:08,056 --> 00:00:11,489
PHISHGUARD AI helps us pause,
inspect the evidence,

4
00:00:11,529 --> 00:00:13,960
and understand the risk before we click.

5
00:00:18,200 --> 00:00:22,433
Imagine a fictional social post asking you
to verify your account.

6
00:00:22,473 --> 00:00:26,527
Its link contains the word Facebook, but
that alone proves nothing.

7
00:00:26,567 --> 00:00:29,544
The real question is which domain
receives your visit,

8
00:00:29,584 --> 00:00:31,554
and what warning signs appear.

9
00:00:36,200 --> 00:00:38,976
Our planned flow begins with Analyze Page.

10
00:00:39,016 --> 00:00:44,166
The extension collects links from the current page
and sends them to a local backend.

11
00:00:44,206 --> 00:00:47,227
URL checks identify structural signals,

12
00:00:47,267 --> 00:00:51,049
then AI helps classify and explain
the evidence.

13
00:00:54,200 --> 00:00:57,842
Meta Model API provides the
explanation layer.

14
00:00:57,882 --> 00:00:58,899
For this project,

15
00:00:58,939 --> 00:01:05,119
Muse Spark reads the supplied URL evidence and
describes the risk in plain language.

16
00:01:05,159 --> 00:01:09,205
It does not need to open the suspicious
destination for this assessment.

17
00:01:12,200 --> 00:01:15,586
Jev, from TypeSafe, provides a
structured decision.

18
00:01:15,626 --> 00:01:16,844
Instead of a paragraph,

19
00:01:16,884 --> 00:01:21,620
it returns a category and probabilities that our
software can use directly.

20
00:01:21,660 --> 00:01:26,796
Meta explains the evidence; Jev gives us a
machine-readable classification.

21
00:01:30,200 --> 00:01:33,093
Here is our simulated social-feed scenario:

22
00:01:33,133 --> 00:01:36,901
an account-verification post with a
Facebook-like address.

23
00:01:36,941 --> 00:01:39,112
This interface is an animation,

24
00:01:39,152 --> 00:01:42,595
not a recording of a completed
Facebook integration.

25
00:01:42,635 --> 00:01:46,784
The reserved example domain represents a
fictional destination.

26
00:01:48,200 --> 00:01:53,017
The animated scan highlights three
signals: an HTTP connection,

27
00:01:53,057 --> 00:01:58,252
a brand name on an unrelated domain, and login
or verification wording.

28
00:01:58,292 --> 00:02:00,457
These signals justify caution.

29
00:02:00,497 --> 00:02:04,059
They do not establish what a real
destination contains.

30
00:02:06,200 --> 00:02:10,349
We tested this fictional URL directly
with both APIs.

31
00:02:10,389 --> 00:02:13,194
Both returned HTTP two hundred.

32
00:02:13,234 --> 00:02:15,022
Jev selected high risk,

33
00:02:15,062 --> 00:02:18,547
with a high-category probability of
ninety-two percent.

34
00:02:18,587 --> 00:02:23,190
Meta also assessed high phishing risk and
explained the domain and wording.

35
00:02:24,200 --> 00:02:28,422
Today, the public website offers
browser-local URL checks,

36
00:02:28,462 --> 00:02:32,339
and separate live tests verify both
API connections.

37
00:02:32,379 --> 00:02:35,331
The extension and backend source
are available.

38
00:02:35,371 --> 00:02:42,072
Connecting the verified Jev format and both models into the
complete app remains work in progress.

39
00:02:42,200 --> 00:02:44,271
PHISHGUARD AI, LOUFFI Edition.

40
00:02:44,311 --> 00:02:46,930
Transparent signals, structured decisions,

41
00:02:46,970 --> 00:02:49,453
and explanations people can understand.

42
00:02:49,493 --> 00:02:51,784
Explore the prototype and source code.

43
00:02:51,824 --> 00:02:52,948
Our goal is simple:

44
00:02:52,988 --> 00:02:56,942
help people make a more informed decision
before trusting a link.