Tobin South's Avatar

Tobin South

@tobinsouth.bsky.social

PhD student @MIT & @medialab; Building a secure & robust computational future; πŸ‡¦πŸ‡ΊπŸ¦˜; Fulbrighter; he/him

63 Followers  |  94 Following  |  15 Posts  |  Joined: 08.02.2024  |  2.0584

Latest posts by tobinsouth.bsky.social on Bluesky

Trying to build some authentication protocols into it in collaboration with a bunch of Stanford folks. Very fun space!

13.02.2025 19:15 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

πŸ“„ Read the full paper (examples, motivations, legal insights, technical designs, and more): arxiv.org/pdf/2501.09674

27.01.2025 17:55 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Amazing to collaborate with Samuele Marro, Thomas Hardjono, Robert Mahari, @cedricwhitney.bsky.social, Dazza Greenwood, Alan Chan, Alex Pentland, and many others providing feedback.

27.01.2025 17:55 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

We’ve even tackled natural language permissions!

Our framework translates human-readable instructions into auditable, structured access controls.

Think: "Only read project files, avoid financial folders" β†’ enforceable, machine-readable rules. πŸš€

27.01.2025 17:55 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Delegation isn't just technical. It's crucial for:

β‹… Protecting human-only spaces online 🌐
β‹… Preventing misuse in high-stakes domains (finance, healthcare) πŸ₯πŸ’³
β‹… Enabling safe, auditable agent collaboration πŸ› οΈ

27.01.2025 17:55 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Our framework builds on existing web standards (OAuth 2.0, OpenID Connect) to:

βœ… Identify agents uniquely
βœ… Limit permissions & scope
βœ… Maintain audit trails
βœ… Ensure human oversight

All while maintaining compatibility with web infrastructure. 🌐

27.01.2025 17:55 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
Post image

Autonomous AI agents are growing fastβ€”how do we ensure they act responsibly, securely, and transparently?

OpenAI Operator isn't good enough...

πŸ”‘ Authenticated Delegation is our answer:
A secure way to delegate authority while maintaining accountability and control.

27.01.2025 17:55 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
Post image

How can we build better delegation to AI agents?

🚨 New Paper on *Autheticated Delegation* for AI Agents. Let’s make human-in-the-loop systems work for people.

27.01.2025 17:55 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 1    πŸ“Œ 1

Thank you @melissahei.bsky.social for covering the Data Provenance Initiative's new study on 🌟 Multimodal data 🌟.

Our study finds:
- A concentration of data power
- Western-centricity isn't improving for 10+ years

Read more! www.dataprovenance.org/Multimodal_D...

18.12.2024 18:36 β€” πŸ‘ 11    πŸ” 6    πŸ’¬ 1    πŸ“Œ 0

At this point, I'm just racing to write my PhD thesis before o4 comes out and would do a better job

20.12.2024 19:25 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

The cursor team really can't catch a break

20.12.2024 19:25 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Post image Post image

@klyman.bsky.social is hands down the funnest Evals talk yet.

15.12.2024 22:55 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

This is the best abstract I've perhaps ever seen...

02.12.2024 23:01 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

I always think not eating until the feast is a smart move until I arrive and someone starts handing me French 75s

29.11.2024 18:31 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

@mbakker.bsky.social is the Troye Sivan of AI

26.11.2024 21:13 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

I'm glad I have access to a new app that I can lurk in and post once a year

24.11.2024 19:02 β€” πŸ‘ 4    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

@tobinsouth is following 19 prominent accounts