If you enjoy this newsletter, please become our paid subscriber to help this keep going.
Highlights
Yoshua Bengio AI safety lab.OpenAI 3rd Software Approach.Trumps say No to AI.Coinbase breach linked to India.Github hacked by Agents.Special:
Deep Dive
LawZero: AI safety lab YoshuaBengio
Founded by Yoshua Bengio.
Little faith on OpenAI and Google.
Non profit AI lab focused on safety.
Safety by design with Scientist AI.
Non-agentic and trustworthy system.
Headsup: Safety by Desgin will be a top focus!
Approach with 3rd software OpenAI
Policy to follow vulnerabilities.
Validate, prioritize, contact, go public.
Developer friendly, open ended by default.
Necessary practices, i.e. find zero day.
In principles and disclosure workflow.
Headsup: Check your 3rd party companies!
Trump: No AI Safety Institute Forbes
New: Center for AI Standards and Innovation.
Innovation over red tape.
Remove censorship and regulation.
Paradoxical phrasing (AISI → CAISI)
NIST decline to comment.
Headsup: Our upcoming review for AI regulation!
CoinBase: Breach link to India Reuters
Employee took photographs on computer.
Caught when using her personal phone.
Breach cost est. $400 millions.
Coinbase blame “support agents overseas”.
May 11: Extortion demand → wider campaign.
Headsup: This is a trend!
Github: Hacked by AI agents InvariantsLab
No hacking tool or breaking tool.
Fake account created a public project.
Inserted instruction to trick Github AI.
AI read the fake issue, follow instruction.
Grabbed private data from other private.
Shared that private info in the public.
Headsup: Consider AI agent policies.
Our upcoming book
Notice: Y2 GenAI Safety and Security is on GumRoad and Amazon with paperback.
Share this post