Cover Image for Frontier Safety Framework 3.0

Presented by

Montréal AI safety, ethics, and governance

Montréal community of researchers, builders, policymakers, and lively persons interested in advancing AI safety, ethics, and governance through talks, workshops, discussions, and working sessions.

Hosted By

2 Going

AI

Frontier Safety Framework 3.0

Name: Frontier Safety Framework 3.0
Start: 2025-11-11T19:00:00.000-05:00
End: 2025-11-11T20:30:00.000-05:00
Location: Register to See Address

Montréal AI safety, ethics, and governance

Register to See Address

Welcome! To join the event, please register below.

You will be asked to verify token ownership with your wallet.

About Event

A focused briefing on DeepMind’s new Frontier Safety Framework 3.0: how Critical Capability Levels (CCLs) flag risky capabilities and how to pair them with security tiers, deployment mitigations, and a safety case to decide if/when to ship.

We’ll cover (with quick examples):

CCLs for CBRN, Cyber, Harmful Manipulation; and ML R&D (acceleration vs automation).
Core mitigations: weight-security (SL2–SL4), safety post-training, monitoring/IR, access controls, jailbreak defense, verification/KYC where appropriate.
Minimal safety case: evidence, scope limits, and a clear “proceed / proceed with limits / hold” decision.

Format (60–75 min): 20-min briefing → 20-min small-group mapping exercise → Q&A.

Presented by

Montréal AI safety, ethics, and governance

Montréal community of researchers, builders, policymakers, and lively persons interested in advancing AI safety, ethics, and governance through talks, workshops, discussions, and working sessions.

Hosted By

2 Going

AI