Google DeepMind Frontier Safety Framework — provisions explorer

A procurement officer evaluating a vendor against Google DeepMind Frontier Safety Framework can use this page to walk the operative articles + paragraphs the instrument touches, anchored to the primary source. Each row groups every coverage cell that names that article so the reader sees WHICH governance topic each clause speaks to. Use it as a checklist scaffold, not as legal advice (see charter §7.4).

Unanchored citations

8 cells

governs
Foundation Models / GPAI
FSF applies to Google DeepMind frontier-model releases
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
governs
Catastrophic & Existential Risk
FSF Critical Capability Levels (CCL) — explicit thresholds for autonomy, biosecurity, cyber, persuasion
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
governs
Agentic AI Governance
FSF Critical Capability Levels — Autonomy is one of four named CCL domains
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
implicit
Transparency Obligations
FSF publication discloses framework + thresholds; per-evaluation outputs not consistently public
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
implicit
International Coordination
Seoul Frontier AI Safety Commitments signatory; UK AISI pre-deployment evaluation cooperation
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
implicit
Open-Weight Frontier Release
Framework applies to Google DeepMind deployments (mostly closed); third-party open release not addressed
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
implicit
AI in Elections
Persuasion is one of four CCL domains; election-AI as the central example
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
implicit
Compute + Model-Weight Export Controls
FSF mitigations include model-weight access controls + restricted-deployment options
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)

Unanchored citations

8 cells

governs
Foundation Models / GPAI
FSF applies to Google DeepMind frontier-model releases
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
governs
Catastrophic & Existential Risk
FSF Critical Capability Levels (CCL) — explicit thresholds for autonomy, biosecurity, cyber, persuasion
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
governs
Agentic AI Governance
FSF Critical Capability Levels — Autonomy is one of four named CCL domains
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
implicit
Transparency Obligations
FSF publication discloses framework + thresholds; per-evaluation outputs not consistently public
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
implicit
International Coordination
Seoul Frontier AI Safety Commitments signatory; UK AISI pre-deployment evaluation cooperation
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
implicit
Open-Weight Frontier Release
Framework applies to Google DeepMind deployments (mostly closed); third-party open release not addressed
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
implicit
AI in Elections
Persuasion is one of four CCL domains; election-AI as the central example
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)
implicit
Compute + Model-Weight Export Controls
FSF mitigations include model-weight access controls + restricted-deployment options
Verbatim text:
Verbatim text not yet transcribed — see primary source.
Open primary source: Google DeepMind Frontier Safety Framework (May 2024)