Skip to main content
v2026.11,986 entries · CC-BY 4.0
CASRAIAsk CASRAI · AI assistant for researchAsk a research question. Answers come with their sources.Funder rules, compliance, methods and publishing. Answers link the sources they rest on.
NIKOLAI elementN9 · Commitments and governanceProposednikolai-v0.3

Framework Update Protocol

Definition

A framework update protocol is the set of rules for revising an AI safety framework: how often it is reviewed, what triggers an early review, who proposes and who approves changes, and how quickly the revised version and its changelog are published.

Also called: Update protocol, Framework change management, Updating the Framework, Policy changes, Updating policies over time

NIKOLAI's proposed operational definition

NIKOLAI proposal: a Framework Update Protocol is an if-then rule set governing when and how a safety framework may be revised -- the declared review cadence, the triggering conditions for an out-of-cycle review, who may propose a change, who must approve it, the publication deadline after approval, and whether a change-log is required. This wording is a NIKOLAI editorial synthesis, not a quotation from any single source.

The operational definition is CASRAI's own proposal, not a definition any named organisation has agreed to. See what NIKOLAI is and is not, or browse the AI safety glossary.

Examples

Examples from published documents

  • "Changes to the RSP will be proposed by the CEO and RSO, and approved by the Board in consultation with the LTBT. If we update the RSP, we will publicly share the updated version prior to or on its effective date and will record the differences from the prior draft in the Change Log." (Anthropic RSP v3.4, section 4(8))
  • "A large frontier developer shall review and, as appropriate, update its frontier AI framework at least once per year." Material modifications must be published with a justification within 30 days (California SB 53, section 22757.12(b))
  • "Any updates to our practices will be reviewed by Microsoft's Chief Responsible AI Officer prior to their adoption. All material revisions will be made public within 30 days of adoption" (Microsoft Frontier Governance Framework, section 5)

Counter-examples

Not to be confused with

  • A framework changelog records what changed; the update protocol is the rule that decides when a change may happen, who approves it and when it must be published.
  • Re-running evaluations on a model after a capability jump reassesses the model under the existing framework; it is not a revision of the framework itself.

Related terms

Related AI safety terms

Terms defined under this element that are not NIKOLAI elements themselves, each as the cited document uses it. Every term is also listed in the AI safety glossary.

Review cadence
Review cadence is how often a developer re-examines its safety framework. The EU GPAI Code of Practice sets a Framework assessment every 12 months, or sooner when there are reasonable grounds to believe the framework's adequacy or adherence has been materially undermined.
Source: EU GPAI Code of Practice, Safety and Security Chapter
Material modification
A material modification is a significant change to a frontier AI framework. Under California SB 53, a large frontier developer that makes one must publish the modified framework and a justification for the change within 30 days.
Source: California SB 53
Living document
A living document is a policy its author expects to keep revising as knowledge improves. Anthropic describes its Responsible Scaling Policy as "a living document" that it will continually update as it learns more about AI capabilities and risks.
Source: Anthropic Responsible Scaling Policy v3.4

Source of record

Where this definition comes from

  • Anthropic Responsible Scaling Policy v3.4, §4(8), pp.16-17

    “Changes to the RSP will be proposed by the CEO and RSO, and approved by the Board in consultation with the LTBT. If we update the RSP, we will publicly share the updated version prior to or on its effective date and will record the differences from the prior draft in the Change Log.”

    https://cdn.sanity.io/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf
  • EU GPAI Code of Practice, Safety and Security Chapter, Commitment 1, Measure 1.3

    “Signatories will update the Framework as appropriate, including without undue delay after a Framework assessment (specified in the following paragraphs), to ensure the information in Measure 1.1 is kept up-to-date and the Framework is at least state-of-the-art. For any update of the Framework, Signatories will include a changelog, describing how and why the Framework has been updated, along with a version number and the date of change.”

    https://ec.europa.eu/newsroom/dae/redirection/document/118119
  • California SB 53, 22757.12(b)(1)

    “A large frontier developer shall review and, as appropriate, update its frontier AI framework at least once per year.”

    https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260SB53

Crosswalk

How named organisations use this concept

Every row below is a shadow mapping. A shadow row is CASRAI's own reading of a published document. No lab, evaluator or regulator named on a shadow row has declared, endorsed, or been consulted on it. That changes only when an organisation files its own Mapping Declaration.
OrganisationTheir term, as publishedMatch & verificationSource
AnthropicShadow mapping
Anthropic RSP v3.4
“§4(8) verbatim: "Policy changes: Changes to the RSP will be proposed by the CEO and RSO, and approved by the Board in consultation with the LTBT. If we update the RSP, we will publicly share the updated version prior to or on its effective date and will record the differences from the prior draft in the Change Log." §4(6)-(7): "Internal review: We will regularly conduct an internal review of our implementation of this policy" and "Procedural compliance review: On approximately an annual basis, we will commission a third-party review that assesses whether we adhered to this policy's main procedural commitments." Nine dated versions from v1.0 (September 19, 2023) to v3.4 (July 8, 2026) in the Changelog; glossary terms "tentative, and they may change in future risk reports" (Risk Report §2.5 fn 12).”
Also cites the Anthropic Risk Report (point-release history) and METR (cross-lab framing).
exactEQ
confidence: high
Anthropic Responsible Scaling Policy v3.4
OpenAIShadow mapping
OpenAI Preparedness Framework v2 / Frontier Governance Framework
“"We will review and potentially update the Preparedness Framework for continued sufficiency at least once a year." (App. B). FGF: "Updates to this FGF may be proposed by Open AI's Safety Advisory Group, Head of Safety Systems, Head of Preparedness, Chief Information Security Officer, Chief Compliance Officer, General Counsel, or executive leadership"; material updates presented to two boards "with changes and justifications for material updates documented in a changelog and published within 30 days of the update" (§7.1); "Framework Assessment" at least every 12 months (§7.2).”
The once-a-year review commitment quoted first is from Preparedness Framework v2, App. B; the proposer/changelog/assessment detail is from the Frontier Governance Framework.
exactEQ
confidence: high
OpenAI Frontier Governance Framework
Google DeepMindShadow mapping
Frontier Safety Framework v3.1
“"The Frontier Safety Framework will be reviewed at least once a year--more frequently if we have reasonable grounds to believe the adequacy of the Framework or our adherence to it has been materially undermined." "The updated version and framework assessment will be reviewed by the appropriate corporate governance bodies." (s.5.1)”exactEQ
confidence: high
Google DeepMind Frontier Safety Framework v3.1
xAIShadow mapping
xAI Frontier AI Framework (30 Jun 2026)
“"We will review this Framework periodically, and expect significant changes as the capabilities of our models expand." (s.1). No change log.”
The xAI framework's PDF metadata reads 'Privileged/Confidential DRAFT working FRAMEWORK DOC'; draft-vs-final status unresolved. Coded DU because the review trigger and any changelog requirement are declared but left undefined beyond 'periodically'.
noneDU
confidence: medium
xAI Frontier AI Framework (30 Jun 2026)
MetaShadow mapping
Meta Advanced AI Scaling Framework v2
“Update "when we have reasonable grounds to believe either its adequacy or our adherence to it has been materially undermined, or at least every twelve months, whichever is sooner"; "have it confirmed by our Director of Alignment and Risk and Chief AI Officer, and publish the updated version" (§5.1).”exactEQ
confidence: high
Meta Advanced AI Scaling Framework v2
EUShadow mapping
EU GPAI Code of Practice, Safety and Security Chapter
“Measure 1.3: Framework updated "as appropriate, including without undue delay after a Framework assessment ... to ensure the information in Measure 1.1 is kept up-to-date and the Framework is at least state-of-the-art", with "a changelog ... along with a version number and the date of change." A Framework assessment runs "if they have reasonable grounds to believe that the adequacy of their Framework and/or their adherence thereto has been or will be materially undermined, or every 12 months starting from their placing of the model on the market, whichever is sooner", and covers both "Framework adequacy" and "Framework adherence" (incl. documented non-adherence and remediation plans). Updated Framework given to the AI Office "within five business days" of confirmation (Measure 1.4).”exactEQ
confidence: high
EU GPAI Code of Practice, Safety and Security Chapter
California SB 53Shadow mapping
California SB 53
“"(1) A large frontier developer shall review and, as appropriate, update its frontier AI framework at least once per year. (2) If a large frontier developer makes a material modification ... publish the modified frontier AI framework and a justification for that modification within 30 days." (22757.12(b))”exactEQ
confidence: high
California SB 53
METRShadow mapping
METR
“"Updating Policies Over Time: Intentions to update the policy periodically, with protocols detailing how this will be done."”exactEQ
confidence: medium
METR
Seoul Summit signatoriesShadow mapping
Frontier AI Safety Commitments (Seoul 2024)
“"Given the evolving state of the science in this area, the undersigned organisations' approaches (as detailed in paragraphs I-VIII) to meeting Outcomes 1, 2 and 3 may evolve in the future. In such instances, organisations will provide transparency on this, including their reasons, through public updates."”closeCL
confidence: medium
Frontier AI Safety Commitments (AI Seoul Summit 2024)
MicrosoftShadow mapping
Microsoft Frontier Governance Framework (Feb 2026)
“"At least every twelve months, we will have an explicit discussion on how this framework may need to be improved." "Any updates to our practices will be reviewed by Microsoft's Chief Responsible AI Officer prior to their adoption. All material revisions will be made public within 30 days of adoption, and, where appropriate, may be published at the same time as we adopt them." (s.5, 'Updates to this framework', p.9)”exactEQ
confidence: high
Microsoft Frontier Governance Framework (Feb 2026)
G42Shadow mapping
G42 Frontier Safety Framework
“"Changes to this Framework will be proposed by the Frontier AI Governance Board and approved by the G42 Executive Leadership Committee." "An annual external review of the Framework will be conducted" (s.5, 'Policy Review').”
Proposer, approver and an annual external review are named; no publication deadline or changelog requirement was found.
exactEQ
confidence: medium
G42 Frontier Safety Framework
AmazonShadow mapping
Amazon Frontier Model Safety Framework
“"We will therefore revisit this Framework at least annually and update it as necessary" (s.4, p.5); "This Framework is a living document and will be updated to reflect evolving model capabilities and advances in the science underlying AI safety evaluations and mitigations." (preamble, p.1)”
CL: cadence stated; no proposer, approver, changelog or publication deadline found.
closeCL
confidence: medium
Amazon Frontier Model Safety Framework
MagicShadow mapping
Magic AGI Readiness Policy
“Threshold and benchmark changes only: "we may update this threshold upward over time. We may also modify the public and private benchmarks used. Such a change will require approval by our Board of Directors, with input from external security and AI safety advisers." (How We Will Update, item 3)”
NR: covers changes to thresholds and benchmarks, not the whole policy. The page carries the banner "Note: This policy is outdated. We're working on an update."
narrowNR
confidence: medium
Magic AGI Readiness Policy
New York (RAISE Act, S8828)Shadow mapping
NY RAISE Act (S8828)
“"A large frontier developer shall review and, as appropriate, update its frontier AI framework at least once per year." (s.1421(2)); material modifications published with justification within 30 days (s.1421(2)).”
Statutory duty from 1 Jan 2027, according to the text of the Act that CASRAI reviewed; wording aligned with California SB 53.
exactEQ
confidence: high
NY RAISE Act (S8828)
US Congress (H.R. 9925, FRONTIER Act, not enacted)Shadow mapping
FRONTIER Act (H.R. 9925)
“"not later than 30 days after the date of such modification, such developer shall publish on a publicly available website a justification for such modification." (Sec. 4(b))”
Introduced bill, not enacted. CL: covers the justification-publication step for material modifications.
closeCL
confidence: medium
FRONTIER Act, 119th Congress (H.R. 9925, introduced, not enacted)
What do these codes mean?
exact
The source term is equivalent to this element
close
The source term is close but not equivalent to this element
broad
The source term is broader than this element
narrow
The source term is narrower than this element
none
No mapping claim — used for false-friend and declared-but-undefined rows
EQ
Equivalent
CL
Close
BR
Source is broader than the element
NR
Source is narrower than the element
FF
False friend — same or similar label, different meaning
DU
Declared but undefined by the source
UV
Unverified

Divergence

Where sources materially disagree

Sources agree on an annual review but differ on who approves a change: the Board in consultation with the Long-Term Benefit Trust (Anthropic); OpenAI's Legal function oversees updates, with material updates presented to two boards for oversight (OpenAI FGF); the Director of Alignment and Risk and Chief AI Officer confirm (Meta); the Chief Responsible AI Officer reviews (Microsoft); the Executive Leadership Committee approves (G42); 'appropriate corporate governance bodies' review (Google DeepMind); no approver is stated (xAI). Publication clocks range from 'prior to or on its effective date' (Anthropic) to 30 days (SB 53, RAISE, OpenAI FGF, Microsoft) and five business days to the EU AI Office.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →