Source of record
Where this definition comes from
OpenAI "Pacing model development for cyber capabilities", August 2026 statement
“"a two-week pause in reinforcement learning (RL) training on our latest models intended for deployment"; "Our largest planned frontier RL run remains on hold."”
https://openai.com/index/pacing-model-development-cyber-capabilities/Meta Advanced AI Scaling Framework v2, §3.3
“"we will only continue development of the Frontier AI if our risk assessments are complete and safeguards are defined, implemented and validated to reduce risk to the moderate or lower risk threshold" (§3.3).”
https://ai.meta.com/static-resource/Meta_Advanced-AI-Scaling-Framework-v2EU GPAI Code of Practice, Safety and Security chapter, Measure 4.2
“Measure 4.2: "Signatories will only proceed with the development, the making available on the market, and/or the use of the model, if the systemic risks stemming from the model are determined to be acceptable."”
https://ec.europa.eu/newsroom/dae/redirection/document/118119
Crosswalk
How named organisations use this concept
| Organisation | Their term, as published | Match | Source |
|---|---|---|---|
| Anthropic Risk Report / Amodei essay / Anthropic Institute post / Advanced AI Framework | “No explicit halt clause in the Risk Report itself; it says "We do not believe the ideal ecosystem would require a pause on developing models like these until they could be kept under even stronger model weight security. This also is not an obvious call" (§4.6.2). Amodei's Global Pacing Level 4: "full pacing, or even 'pause'". AAF remedy: "Prohibitions on deployment of further covered models until the developer corrects the violations" (p.11). Anthropic Institute post describes Anthropic expecting it "would slow down or temporarily pause, if other developers at or near the frontier also did so in a verifiable manner," and that "a credible pause also has to specify what triggers it, what lifts it, and who adjudicates."” The Anthropic Institute conditional-pause wording is flagged [VERIFY] in the source document's own open register (item 15: "Anthropic Institute conditional pause exact wording {RSI}") — not yet confirmed against the live page at the time of this crosswalk pass; treat the quoted phrasing as provisional. | narrow confidence: medium | Anthropic Risk Report, August 2026 |
| OpenAI Preparedness Framework v2 / "Pacing model development for cyber capabilities" / "An Alien Mind" | “PF Critical: "halt further development" until Critical-standard safeguards and security controls are specified (Table 1). August 2026: "a two-week pause in reinforcement learning (RL) training on our latest models intended for deployment"; "Our largest planned frontier RL run remains on hold." Operational stop rule: "If they cannot conclusively determine within 30 minutes that the flag is a false positive, those teams are expected to pause the activity." Jakub Pachocki (OpenAI Chief Scientist), "An Alien Mind": "OpenAI will continue to seek technical solutions to alignment and monitoring, to build defensive systems and unilaterally withhold further scaling as needed; however, I believe broader interventions are required" — no trigger, threshold, evidence standard or named decision-maker specified for this conditional commitment.” | exact confidence: high | OpenAI Preparedness Framework v2 |
| xAI xAI Frontier AI Framework (30 Jun 2026) | “"xAI will only proceed with the development, the making available on the market, and/or the use of the model, if the systemic risks stemming from the model are determined to be acceptable." (s.2.3). "we may temporarily fully shut down the relevant system until we have developed a more targeted response" (s.3).” PDF metadata /Title reads "Privileged/Confidential DRAFT working FRAMEWORK DOC"; no xAI statement disambiguating draft vs. final was found. | close confidence: high | xAI Frontier AI Framework (30 Jun 2026, draft-marked) |
| Meta Meta Advanced AI Scaling Framework v2 | “"we will only continue development of the Frontier AI if our risk assessments are complete and safeguards are defined, implemented and validated to reduce risk to the moderate or lower risk threshold" (§3.3). Change log: "Critical threshold changed from 'Stop' to 'Develop with Mitigations.'"” | close confidence: high | Meta Advanced AI Scaling Framework v2 |
| EU EU GPAI Code of Practice, Safety and Security chapter | “Measure 4.2: "Signatories will only proceed with the development, the making available on the market, and/or the use of the model, if the systemic risks stemming from the model are determined to be acceptable." If not acceptable: "(1) not make the model available on the market, restrict the making available on the market ..., withdraw, or recall the model, as necessary; (2) implement safety and/or security mitigations ...; and (3) conduct another round of" risk identification, analysis and acceptance determination. No named authority to lift a halt beyond re-running the process.” | exact confidence: high | EU GPAI Code of Practice, Safety and Security chapter |
| METR METR Common Elements | “"Conditions for Halting Deployment Plans: Commitments to stop deploying models if the AI capabilities of concern emerge before the appropriate mitigations can be put in place." "Conditions for Halting Development Plans: Commitments to halt model development if capabilities of concern emerge before the appropriate mitigations can be put in place."” | exact confidence: high | METR (common-elements) |
| Frontier Model Forum FMF Risk Taxonomy and Thresholds | “Frameworks "stipulate clear thresholds at which deployment or further development of a model may be paused" (s1.1).” | close confidence: high | FMF Risk Taxonomy and Thresholds |
| Safety Framework Cards (discovery) Safety Framework Cards (SSRN 7061798) | “"governance triggers" dimension named in discovery sweep; full text paywalled/unread.” Unverified: full text inaccessible to the source document's own research pass. | none confidence: low | Discovery sweep — Safety Framework Cards (SSRN 7061798, paywalled/unread) |
| Microsoft Microsoft Frontier Governance Framework (Feb 2026) | “"If ... we identify a risk we cannot sufficiently mitigate, we will pause development and deployment" (Sec. 4, p.8).” | exact confidence: high | Microsoft Frontier Governance Framework (Feb 2026) |
| G42 G42 Frontier Safety Framework | “G42 pauses capability development if the required Security Mitigation Level cannot be met.” | exact confidence: medium | G42 Frontier Safety Framework |
| Magic Magic AGI Readiness Policy | “Magic halts further model development until dangerous-capability evaluations are ready (see B8 quote).” | exact confidence: medium | Magic AGI Readiness Policy |
Related, not mapped
Pointers that are not crosswalk claims
These sources mention this concept but do not define or map it clearly enough to count as a crosswalk row — noted here so the research is visible without overstating it as a mapping.
- US EO 14409 / UK Seoul Commitments (GOV)
EO 14409 excludes any "mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models" (Sec. 3(c)); Seoul Commitment IV: "In the extreme, organisations commit not to develop or deploy a model or system at all, if mitigations cannot be applied to keep risks below the thresholds." Related governance commitments, not a stated halt-condition rule with trigger/lift authority. Scored RL in the source document.
Executive Order 14409 / Seoul Frontier AI Safety Commitments
Gap
Only OpenAI's August 2026 actions show a halt actually invoked, and no source names who may lift one or on what evidence ("no resumption criteria" {OPACE}).
