Safety frameworks cluster: Constitutional AI v3 + Preparedness v2 + Frontier Safety v3 + Llama Guard 4 + AI Red Team + C2PA 2.0 — every frontier lab ships a safety framework update

8 sources

The week of 2026-08-05 saw a cluster of safety and governance announcements: Anthropic Constitutional AI v3, OpenAI Preparedness Framework v2, Google DeepMind Frontier Safety Framework v3, Meta Llama Guard 4 (open-weight safety classifier), Microsoft AI Red Team updates + AI Risk Shadow Model, C2PA Content Credentials 2.0 (cryptographic media provenance). The Decoder frames the cluster as 'every frontier lab ships a safety framework update'. The pattern: every frontier lab is publishing / updating its safety framework at the same cadence as its model releases, and the open-weight tooling (Llama Guard, C2PA) is shipping in lockstep.

Why now

The cluster is the editorial framing that turns six independent safety announcements into a single coherent 'safety frameworks are now part of the model release cadence' story for creators — useful because the comparison is most informative when shown side-by-side.

Why it is worth publishing

Demo potential: a side-by-side of Constitutional AI v3 vs Preparedness v2 vs Frontier Safety v3 vs Llama Guard 4 vs AI Red Team vs C2PA 2.0 — what each framework actually constrains and what each creator-facing surface they expose.

Evidence basis

The Decoder + IT之家 weekly roundups + six independent lab primary sources

“Six frontier safety frameworks shipped in one week — Constitutional AI v3, Preparedness v2, Frontier Safety v3, Llama Guard 4, AI Red Team, C2PA 2.0 — and every frontier lab is now updating its safety framework at the same cadence as its models.”

Angle

Use the cluster to introduce the 'safety frameworks are now part of the model release cadence' pattern — every frontier lab publishes / updates its safety framework at the same cadence as its models — and use that lens to compare frameworks side-by-side.

Format

Long-form explainer

Demo idea

Record a 16-minute comparison explainer: 2 min intro on 'safety frameworks at model release cadence' framing, then 2 min per framework (Constitutional AI / Preparedness / Frontier Safety / Llama Guard / AI Red Team / C2PA), then a 4-min side-by-side of what each framework actually constrains and what each creator-facing surface they expose.

Platform notes

Each lab frames its release against the competitive set it cares about; The Decoder and IT之家 are editorial framing layers, not independent verification. Specific principle lists, risk categories, score thresholds, and early-warning thresholds beyond the captured summary were not extracted; confirm any specific number or category name against the underlying lab docs before stating it on the record.

Usable claims

Anthropic released Constitutional AI v3 specification — multi-stage training with explicit principles, harmlessness reward model + helpfulness reward model, and public critique-revision loop.
OpenAI updated the Preparedness Framework to v2 — risk categories, score thresholds, mitigation obligations, and cross-functional review board.
Google DeepMind released Frontier Safety Framework v3 — capability evaluations, early-warning indicators, mitigation deployment obligations.
Meta released Llama Guard 4 open-weight safety classifier — multi-class taxonomy for unsafe content, integration with Llama Stack 1.5 reference server.
Microsoft updated AI Red Team methodology and AI Risk Shadow Model framework for internal red-team tracking.
C2PA released Content Credentials 2.0 specification — cryptographic provenance for media assets, model + tool attestations, tamper-evident manifests.

Evidence pipeline

From the news

Breakdown

Six frontier safety frameworks shipped in one week — the editorial framing ('every frontier lab ships a safety framework update') is useful but turns into a lab marketing release if you don't introduce the cadence pattern. This explainer uses the cluster to introduce the 'safety frameworks at model release cadence' pattern and uses that lens to compare frameworks side-by-side, while keeping the per-framework structure honest (Constitutional AI works at training time; Preparedness v2 / Frontier Safety v3 are evaluation frameworks; Llama Guard 4 + C2PA 2.0 are creator-facing surfaces; AI Red Team is post-deployment).

Sources

Risks

Use The Decoder and IT之家 as media-type corroboration, but read the underlying lab docs for any specific safety claim before stating it on the record. Verify specific capability claim against the underlying vendor docs and the actual license / pricing matrix before stating it on the record; do not paraphrase per-platform pricing or license terms into specific dollar figures or commercial-use clauses.
Docs confirm the existence of the framework versions but specific thresholds, categories, and tasks beyond the captured summary are not in this pass. Verify specific capability claim against the underlying vendor docs and the actual license / pricing matrix before stating it on the record; do not paraphrase per-platform pricing or license terms into specific dollar figures or commercial-use clauses.

Demo ideas

Side-by-side comparison: same prompt run through Constitutional AI v3 vs Preparedness v2 vs Frontier Safety v3 vs Llama Guard 4 — what each framework constrains, what each creator-facing surface exposes.
Decision tree: 'which safety framework for which creator use case' (input/output filtering → Llama Guard 4, media provenance → C2PA 2.0, internal evaluation → Preparedness v2 / Frontier Safety v3, training-time safety → Constitutional AI v3, internal red-team → AI Red Team).
Cadence graphic: plot each lab's framework updates against its model releases on a single timeline.