Documents

8 documents, published 20 Aug 2025 to 21 Sep 2026. The chart runs from each document's publication to its latest known version; the table lists them newest first, each linking to its page.

xAI's documents and the versions we know of

One row per document, from its publication to its latest dated version. A tick marks each later version whose date we know, from a changelog, a date in the document or our source registry.

Jul 2025Oct 2025Jan 2026Apr 2026Jul 2026Oct 2026Grok 4Grok Code Fast 1Grok 4 FastGrok 4.1Grok 4.20Grok 4.5Grok 4.6Grok 4.7

A copy we retrieved without knowing when that version was published has no tick: its only date is the day we read it. The table lists every version, with how its date is known. Versions known only from a changelog have no copy we hold yet.

Source: Safety Card Ledger v0.1 · Data from developer system cards · CC BY 4.0

xAI's documents, newest first.
DocumentPublishedKindVersions knownChangelogValues
Grok 4.7 Model CardModel card · 1 version known · No changelog · 11 values21 Sep 2026Model card1No changelog11
Grok 4.6 Model CardModel card · 2 versions known · Partial changelog · 22 values12 Aug 2026Model card2Partial changelog22
Grok 4.5 Model CardModel card · 2 versions known · No changelog · 15 values14 Jul 2026Model card2No changelog15
Grok 4.20 System CardSystem card · 1 version known · No changelog · 18 values7 Apr 2026System card1No changelog18
Grok 4.1 Model CardModel card · 1 version known · No changelog · 14 values17 Nov 2025Model card1No changelog14
Grok 4 Fast Model CardModel card · 1 version known · No changelog · 10 values19 Sep 2025Model card1No changelog10
Grok Code Fast 1 Model CardModel card · 1 version known · No changelog · 6 values26 Aug 2025Model card1No changelog6
Grok 4 Model CardModel card · 3 versions known · No changelog · 17 values20 Aug 2025Model card3No changelog17

Frameworks and versions used

A risk determination is a developer's formal decision on whether a model crosses a risk threshold (Methodology §8). They are counted here by the framework and version each one names, as recorded; each model's page shows the levels as printed. Levels are never mapped between frameworks.

4 determinations are recorded from xAI's documents. 3 of them name no framework.

Frameworks and versions named in xAI's risk determinations, with how many determinations name each.
FrameworkVersion namedDeterminationsModelsIn documents
FAIFVersion: Not stated · 1 determination · 1 model · 1 document, 21 Sep 2026Not stated1121 Sep 2026 · 1 document
No framework named3 determinations · 3 models · 3 documents, 20 Aug 2025 to 14 Jul 2026—3320 Aug 2025 to 14 Jul 2026 · 3 documents

Metric families reported

xAI's documents have a printed number in 8 of the 12 metric families. None has one in Reward hacking, Misalignment audits, Self-preservation and Chain-of-thought monitorability.

Metric families in xAI's documents

Share of xAI's 8 documents with at least one printed number in each metric family. A cell opens the data explorer at that family, filtered to xAI.

Share of each developer's documents in the version 0 dataset with at least one numeric value in each metric family.
DeveloperM1M2M3M4M5M6M7M8M9M10M11M12
xAI (8)13%0%13%0%100%88%100%88%63%100%0%0%

Share of documents0%1-24%25-49%50-74%75-100%

M1
Evaluation awareness
M2
Reward hacking
M3
Sabotage and sandbagging
M4
Misalignment audits
M5
Honesty and hallucination
M6
Sycophancy
M7
Harmful compliance and over-refusal
M8
Jailbreak robustness
M9
Prompt injection
M10
Dangerous capabilities and risk determinations
M11
Self-preservation
M12
Chain-of-thought monitorability

Read this as a lower bound: some documents were not read to the end in version 0 (each document's page says how far), so families reported late in them may be undercounted. Values read off a figure, stated in words, or given as a category do not count.

Source: Safety Card Ledger v0.1 · Data from developer system cards · CC BY 4.0

Changelog practice

A document has a changelog when it keeps its own record of the changes made to it after publication. Where it keeps none, changes can only be found by comparing its versions (Methodology §7).

Of xAI's 8 documents, 1 has a partial changelog and 7 have no changelog.

How many of xAI's documents keep a changelog.
ChangelogDocumentsShare
Partial changelogThe document records some of its changes, but not all of them, or not in full.113%
No changelogThe document keeps no record of its changes.788%

Revisions

5 revisions are recorded in 1 document, newest first. Each shows what changed, whether the document's own changelog explains it, and how we know. A revision is not in itself evidence of wrongdoing; most are corrections.

KeyVerifiedUnverifiedDisputedCorrected read off a figure or stated in wordsA value opens its source and history.

  1. xAI · 12 Aug 2026 to 17 Aug 2026

    Grok 4.6 Model Card

    Bio capability lift statement relative to Grok 4.5, Grok 4.6

    M10 · Dangerous capabilities and risk determinations

    no appreciable liftnoted in the biological domain but is limited

    Not explained — The changelog does not mention this change.

    Old value: New value:

    How we know The old value was read in the earlier copy and the new one in the later copy. The FORTRESS-RN addition in the same revision is recorded separately.

    Checked Not yet checked

  2. xAI · 12 Aug 2026 to 17 Aug 2026

    Grok 4.6 Model Card

    FORTRESS-RN R/N refusal accuracy added, Grok 4.6 (high)

    M7 · Harmful compliance and over-refusal

    97.9%

    Not explained — Added without a changelog note.

    New value:

    How we know The value was read in the later copy; the earlier copy does not have it.

    Checked Not yet checked

  3. xAI · 12 Aug 2026 to 17 Aug 2026

    Grok 4.6 Model Card

    HackerBench v0.2 harmful/dual-use compliance, Grok 4.6 (high)

    M7 · Harmful compliance and over-refusal

    16.7%6.9%

    Partly explained — changelog lists "corrected eval results" for this eval; no reason

    Old value: New value:

    How we know The old value was read in the earlier copy and the new one in the later copy. Blind check 26 Sep 2026 (both versions)

    Checked Blind-verified

  4. xAI · 12 Aug 2026 to 17 Aug 2026

    Grok 4.6 Model Card

    MASK dishonesty, Grok 4.6 (high)

    M5 · Honesty and hallucination

    3.8%1.90%

    Partly explained — listed as corrected; no reason

    Old value: New value:

    How we know The old value was read in the earlier copy and the new one in the later copy. Blind check 26 Sep 2026 (both versions)

    Checked Blind-verified

  5. xAI · 12 Aug 2026 to 17 Aug 2026

    Grok 4.6 Model Card

    Self-harm compliance, Grok 4.6 (high)

    M7 · Harmful compliance and over-refusal

    3.7%0.84%

    Partly explained — listed as corrected; no reason

    Old value: New value:

    How we know The old value was read in the earlier copy and the new one in the later copy. Blind check 26 Sep 2026 (both versions)

    Checked Blind-verified

Models

8 models, newest first by the date of each model's first card.

xAI's models, newest first by the date of each model's first card.
ModelFirst cardAvailabilityValues
Grok 4.7First card: 21 Sep 2026 · Public · 10 values21 Sep 2026Public10
Grok 4.6First card: 12 Aug 2026 · Public · 22 values12 Aug 2026Public22
Grok 4.5First card: 14 Jul 2026 · Public · 14 values14 Jul 2026Public14
Grok 4.20First card: 7 Apr 2026 · Public · 16 values7 Apr 2026Public16
Grok 4.1First card: 17 Nov 2025 · Public · 15 values17 Nov 2025Public15
Grok 4 FastFirst card: 19 Sep 2025 · Public · 10 values19 Sep 2025Public10
Grok Code Fast 1First card: 26 Aug 2025 · Public · 6 values26 Aug 2025Public6
Grok 4First card: 20 Aug 2025 · Public · 18 values20 Aug 2025Public18

Not recorded

No restated values are recorded: no later xAI document is recorded as reporting an earlier model's result again.