doi.bio/ariel_herbert-voss


Ariel Herbert-Voss

Ariel Herbert-Voss (also known as Ari Herbert-Voss) is a research scientist at OpenAI and a PhD student in computer science at Harvard University. They hold affiliate researcher positions at the Vector Institute for Artificial Intelligence in Toronto and the MIT Media Lab. Herbert-Voss is interested in the malicious uses and abuses of AI and has conducted research on data exfiltration and the exploitation of AI systems.

Career

Herbert-Voss is a Senior Research Scientist at OpenAI, where they work on breaking machine learning systems at the algorithmic level. They are also a PhD student at Harvard University, where they research vulnerabilities and develop threat models for AI-based systems to aid security researchers in red-teaming these systems. Additionally, they periodically conduct trainings on practical adversarial machine learning.

Herbert-Voss has spoken at Black Hat, DEF CON, and NeurIPS, and is the co-founder and co-organizer of the DEF CON AI Village, a hacker community focused on communicating the uses and abuses of artificial intelligence technology. They are also a non-resident research fellow with the Belfer Center Cyber Project and a fellow at the Berkman-Klein Center for Internet and Society.

Publications

- Don’t Red Team AI Like a Chump

Ariel Herbert-Voss

Ariel Herbert-Voss (also known as Ari Herbert-Voss) is a researcher with a focus on AI security and ethics. They are currently a Senior Research Scientist at OpenAI and a PhD candidate in computer science at Harvard University.

Career

Herbert-Voss is a research fellow with the Belfer Center Cyber Project and holds affiliate researcher positions at the Vector Institute for Artificial Intelligence and the MIT Media Lab. They are also a fellow at the Berkman Klein Center, where they research vulnerabilities and develop threat models for AI-based systems to aid security researchers in red-teaming these systems.

Herbert-Voss has spoken at Black Hat, DEF CON, and NeurIPS, and is the co-founder and co-organizer of the DEF CON AI Village, a community focused on communicating the uses and abuses of AI technology.

Research Focus

Their research focuses on data exfiltration and the exploitation of AI systems, as well as adversary threat modeling, to inform policymakers about the security implications of AI research. They have discovered several exploits in popular AI-based technology products.

Notable Works

Youtube Videos

Youtube Title: Ariel Herbert Voss - Dont Red Team AI Like a Chump - DEF CON 27 Conference

Youtube Link: link

Youtube Channel Name: DEFCONConference

Youtube Channel Link: https://www.youtube.com/@DEFCONConference

Ariel Herbert Voss - Dont Red Team AI Like a Chump - DEF CON 27 Conference

Youtube Title: Graph the Planet 2024 - Interview w/ Ariel Herbert-Voss of RunSybil

Youtube Link: link

Youtube Channel Name: Graphistry

Youtube Channel Link: https://www.youtube.com/@graphistryhome

Graph the Planet 2024 - Interview w/ Ariel Herbert-Voss of RunSybil

Youtube Title: DEF CON 26 AI VILLAGE - Ariel Herbert Voss - Machine Learning Model Hardening For Fun and Profit

Youtube Link: link

Youtube Channel Name: DEFCONConference

Youtube Channel Link: https://www.youtube.com/@DEFCONConference

DEF CON 26 AI VILLAGE - Ariel Herbert Voss - Machine Learning Model Hardening For Fun and Profit

Youtube Title: How NOT to Train Your Hack Bot: Dos and Don'ts of Building Offensive GPTs

Youtube Link: link

Youtube Channel Name: Black Hat

Youtube Channel Link: https://www.youtube.com/@BlackHatOfficialYT

How NOT to Train Your Hack Bot: Dos and Don'ts of Building Offensive GPTs

Youtube Title: Forward Focus: Perspectives on AI, Hype, and Security

Youtube Link: link

Youtube Channel Name: Black Hat

Youtube Channel Link: https://www.youtube.com/@BlackHatOfficialYT

Forward Focus: Perspectives on AI, Hype, and Security

Youtube Title: DEF CON 27 - Ariel Herbert-Voss - Dont Red-Team AI Like a Chump

Youtube Link: link

Youtube Channel Name: HackersOnBoard

Youtube Channel Link: https://www.youtube.com/@HackersOnBoard

DEF CON 27 - Ariel Herbert-Voss - Dont Red-Team AI Like a Chump

Youtube Title: Herbert Voss e il Gargano: presenza e memoria

Youtube Link: link

Youtube Channel Name: Fondazione dei Monti Uniti di Foggia

Youtube Channel Link: https://www.youtube.com/@FondazioneMontiUnitidiFoggia

Herbert Voss e il Gargano: presenza e memoria

Youtube Title: Practical Defenses Against Adversarial Machine Learning

Youtube Link: link

Youtube Channel Name: Black Hat

Youtube Channel Link: https://www.youtube.com/@BlackHatOfficialYT

Practical Defenses Against Adversarial Machine Learning

Youtube Title: USENIX Security '21 - Extracting Training Data from Large Language Models

Youtube Link: link

Youtube Channel Name: USENIX

Youtube Channel Link: https://www.youtube.com/@UsenixOrg

USENIX Security '21 - Extracting Training Data from Large Language Models

Youtube Title: GPT-3: Language Models are Few-Shot Learners (Paper Explained)

Youtube Link: link

Youtube Channel Name: Yannic Kilcher

Youtube Channel Link: https://www.youtube.com/@YannicKilcher

GPT-3: Language Models are Few-Shot Learners (Paper Explained)

Youtube Title: DEF CON 25 - Hyrum Anderson - Evading next gen AV using AI

Youtube Link: link

Youtube Channel Name: DEFCONConference

Youtube Channel Link: https://www.youtube.com/@DEFCONConference

DEF CON 25 - Hyrum Anderson - Evading next gen AV using AI

Youtube Title: A.I. Talks with Animals

Youtube Link: link

Youtube Channel Name: A.I. Talks

Youtube Channel Link: https://www.youtube.com/@aitalks

A.I. Talks with Animals

Youtube Title: Learning Visuomotor Policies for Aerial Navigation Using Cross-Modal Representations

Youtube Link: link

Youtube Channel Name: Microsoft Research

Youtube Channel Link: https://www.youtube.com/@MicrosoftResearch

Learning Visuomotor Policies for Aerial Navigation Using Cross-Modal Representations

Youtube Title: Extracting Training Data from Large Language Models (Paper Explained)

Youtube Link: link

Youtube Channel Name: Yannic Kilcher

Youtube Channel Link: https://www.youtube.com/@YannicKilcher

Extracting Training Data from Large Language Models (Paper Explained)

Youtube Title: Writer CEO on Generative AI and the Enterprise

Youtube Link: link

Youtube Channel Name: Bloomberg Technology

Youtube Channel Link: https://www.youtube.com/@BloombergTechnology

Writer CEO on Generative AI and the Enterprise