2024-02-14 DeepfakesIDV DG Meeting Notes DRAFT

Ajay Amlani

N/A

Andrew Hughes

FaceTec, Inc.

Present

Ashok Singal

Demystify Biometrics

Chris LaBarbera

Verizon

Denny Prvu

RBC Royal Bank of Canada

Present

Iain Corby

SafetyTech Limited

Present

Jay Meier

FaceTec, Inc.

Present

Jim Pasquale

Chasm Crossing Solutions

Jordan Burris

Socure

Justin Byrd

N/A

Present

Maxine Most

Acuity Market Intelligence

Present

Mike Magrath

Easy Dynamics

Peter Davis

Airside Mobile, Inc

Present

Sean Lanzner

N/A

Present

Anand Kumar

SecureKloud Technologies, Inc.

Present

Dan Bachenheimer

Accenture

Present

Dawid Jacobs

DAL Identity

Present

Leonardo Maldonado

GSE

Present

Michael Choudoin

N/A

Present

Michael Engle

1Kosmos

Present

Uttam Reddi

Aware Corp

Present

Kay Chopard

Kantara Initiative

Present

Joey Pritikin

Paravision

Present

Ronald Chapman

iProov

Present

Daniel Molina

iProov

Present

Re-categorization of the content

Jay

• Consider categorization Policy, Provenance (e.g. C2PA Overview - C2PA or CAI), Protection (includes Detection)

• Consider separation of Recognition v Analysis

• Joey to think about drafting material about ‘policy’ also how ‘deepfakes’ intersect with discussions about content authenticity and public discourse

• Consider policy as a means of attack mitigation e.g. on sensor control

• Provenance and deepfakes - e.g. camera sensors could sign data streams as they are originated e.g. OpenAI is adding new watermarks to DALL-E 3

• Agreement to restructure

iProov Taxonomy of Threats 2024 report

Ronald Chapman, Daniel Molina

@ iProov

• Attacks today are novel - the attacks are evolving

• Face Swap tools are widely available and cheap/free

  • This seems to be the preferred method for attackers

• Starting to see real stories of deepfake based attacks

• Starting to see new threat actor groups - doing proof of concept and selling tools online

  • Targeted attacks

• More use of emulators

• Web browsers in mobile apps are a big attack surface

• Contact Sean Lanzer for deeper discussions

  • iProov is willing to demo actual tools and attacks

• Very high definition screens are being used to present face swaps

  • a hybrid attack is more effective right now

• How can a vendor know that a fake is being presented?

  • Device SDK to help protect the OS and sensor - to detect/prevent injections

  • Challenge-response techniques

Comment:

MORE RESEARCH... Not sure if the group has seen this paper from our partners on www.ProjectDefAI.com - Vulnerability of Face Recognition to Deep Morphing" by Pavel Korshunov and Sebastien Marcel ´ Idiap Research Institute, Martigny, Switzerland

"We show that the state of the art face
recognition systems based on VGG and Facenet neural networks are vulnerable to the deep morph videos, with 85.62% and 95.00% false acceptance rates, respectively"

https://defaiproject.com/

Call for contributors

Chairs

Please volunteer for content areas by email

IDPV Core topic area

Andrew Hughes

Andrew walked through the “IDPV Placemat” and solicited comments and feedback.

Please discuss!