Morrigan (formerly Surf-Shield) has come a long way.
Really glad to see the localization improvements — I tested
it with scam messages written in Tagalog and Hiligaynon,
including a fake 4Ps GCash downpayment scam, and it correctly
flagged both as highly suspicious. It even understood regional
context like "buwas" (tomorrow) as an urgency tactic, which
was impressive for an on-device model.
Also tested URL phishing detection with obfuscated links and
it performed well — breaking down suspicious subdomains,
mismatched domains, and generic IDs clearly.
The fake image detection is a huge improvement. We discussed
this limitation previously, and it's great to see Gemini Nano
now handling manipulated or text-overlaid images.
I haven't tested audio call detection in Filipino languages
yet, but plan to.
Morrigan (formerly Surf-Shield) has come a long way.
Really glad to see the localization improvements — I tested
it with scam messages written in Tagalog and Hiligaynon,
including a fake 4Ps GCash downpayment scam, and it correctly
flagged both as highly suspicious. It even understood regional
context like "buwas" (tomorrow) as an urgency tactic, which
was impressive for an on-device model.
Also tested URL phishing detection with obfuscated links and
it performed well — breaking down suspicious subdomains,
mismatched domains, and generic IDs clearly.
The fake image detection is a huge improvement. We discussed
this limitation previously, and it's great to see Gemini Nano
now handling manipulated or text-overlaid images.
I haven't tested audio call detection in Filipino languages
yet, but plan to.