Claim and evidence Close ×
Grok 4 Model Card states evaluation result ↓ The card states that the area of highest concern is Grok 4's expert-level biology capabilities, which it says significantly exceed human expert baselines, while stating that the evaluations measure dual-use knowledge and that a high score does not necessarily mean increased risk. The Grok 4 Model Card states that the area of highest concern is the model's expert-level biology capabilities, which significantly exceed human expert baselines on the benchmarks used, while stating that these evaluations measure dual-use knowledge and that a high score indicates greater capability to enable weapons development, not necessarily increased risk. Copy claim link
Copy claim reference
Machine-checked, not reviewed by an editor Stated in source Supported
Supporting evidence: primary · 1 source; Context: primary · 1 source
Subject version The expression stating 'Last updated: August 20, 2025' · Publication date unknown · Effective date not established · Revised 2025-08-20 Date 2025-08-20 · revision Evaluated system Grok 4 API and Grok 4 Web, with safeguards removed for the dual-use capability assessment What the report measured Dual-use chemical and biological knowledge, measured on public benchmarks By what kind of method Benchmark scoring against stated human expert baselines, with safeguards removed for the assessment The result, in the report’s words The card states that biology capabilities are the area of highest concern and significantly exceed human expert baselines, and that chemistry capabilities are strong The caveats the report gives The card states that these evaluations measure dual-use knowledge, so a high score indicates greater capability to enable weapons development and not necessarily increased risk; that radiological and nuclear capabilities are not evaluated; and that, given existing nonproliferation regimes, it assesses its models as generally posing a low risk of enabling malicious use. Scope and limits A benchmark result and its reading as the developer states them, for the model versions and public benchmarks the card names. It is not a finding about real-world capability or risk, and the card itself says the scores measure dual-use knowledge rather than risk. Safeguards were removed for these measurements, as the card states. Benchmark names and scores are not reproduced beyond the card's own summary sentences; no task, protocol or agent is recorded.
The evidence 03 primary source Supports the claim
Grok 4 Model Card ↗ xAI · Publication date unknown
Copy source reference
Cite the original source Cite the original document, not Selekoa. Check these details against it.
Copy APA
Download BibTeX Download RIS Download CSL-JSON
Source excerpt “The area of highest concern is Grok 4's expert-level biology capabilities, which significantly exceed human expert baselines.” Location in source PDF p. 2, §1 Introduction, summary paragraph
How it was read Retrieved directly from the publisher
Machine-checked · 2026-10-30
Source details and check notes Authors xAI Version Retrieved for the candidate pack on 2026-10-28 (SHA-256 54068c0d9570d94dfc63c08247707b8c48045a79b5bbe865af709af95ee14415). PDF stating 'Last updated: August 20, 2025', served from data.x.ai on 2026-10-28; that stated last update is the expression checked, and the card does not say what any update changed or when it was first published. The x.ai web page for the card returned HTTP 404. Retrieved again at admission on 2026-10-30 (SHA-256 54068c0d9570d94dfc63c08247707b8c48045a79b5bbe865af709af95ee14415, the same bytes). Canonical address Open the canonical source ↗ Accessed 2026-10-28 Language English (en) Attribution xAI, “Grok 4 Model Card”. Reuse No open redistribution licence asserted. Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution. Check notes Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (17 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval whose SHA-256 equals the pack’s. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. A cross-family machine check (2026-10-30, item 39) re-read the record’s sources and found full support. No human editorial review. primary source Context only
Grok 4 Model Card ↗ xAI · Publication date unknown
Copy source reference
Cite the original source Cite the original document, not Selekoa. Check these details against it.
Copy APA
Download BibTeX Download RIS Download CSL-JSON
Source excerpt “Note that these evaluations measure dual-use knowledge: a high score indicates greater capability to enable weapons development, not necessarily increased risk.” Location in source PDF p. 6, §2.3.2 Results
How it was read Retrieved directly from the publisher
Machine-checked · 2026-10-30
Source details and check notes Authors xAI Version Retrieved for the candidate pack on 2026-10-28 (SHA-256 54068c0d9570d94dfc63c08247707b8c48045a79b5bbe865af709af95ee14415). PDF stating 'Last updated: August 20, 2025', served from data.x.ai on 2026-10-28; that stated last update is the expression checked, and the card does not say what any update changed or when it was first published. The x.ai web page for the card returned HTTP 404. Retrieved again at admission on 2026-10-30 (SHA-256 54068c0d9570d94dfc63c08247707b8c48045a79b5bbe865af709af95ee14415, the same bytes). Canonical address Open the canonical source ↗ Accessed 2026-10-28 Language English (en) Attribution xAI, “Grok 4 Model Card”. Reuse No open redistribution licence asserted. Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution. Check notes Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (21 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval whose SHA-256 equals the pack’s. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. A cross-family machine check (2026-10-30, item 39) re-read the record’s sources and found full support. No human editorial review. primary source Context only
Grok 4 Model Card ↗ xAI · Publication date unknown
Copy source reference
Cite the original source Cite the original document, not Selekoa. Check these details against it.
Copy APA
Download BibTeX Download RIS Download CSL-JSON
Source excerpt “We do not evaluate radiological or nuclear capabilities.” Location in source PDF p. 2, §1 Introduction, summary paragraph
How it was read Retrieved directly from the publisher
Machine-checked · 2026-10-30
Source details and check notes Authors xAI Version Retrieved for the candidate pack on 2026-10-28 (SHA-256 54068c0d9570d94dfc63c08247707b8c48045a79b5bbe865af709af95ee14415). PDF stating 'Last updated: August 20, 2025', served from data.x.ai on 2026-10-28; that stated last update is the expression checked, and the card does not say what any update changed or when it was first published. The x.ai web page for the card returned HTTP 404. Retrieved again at admission on 2026-10-30 (SHA-256 54068c0d9570d94dfc63c08247707b8c48045a79b5bbe865af709af95ee14415, the same bytes). Canonical address Open the canonical source ↗ Accessed 2026-10-28 Language English (en) Attribution xAI, “Grok 4 Model Card”. Reuse No open redistribution licence asserted. Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution. Check notes Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (8 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval whose SHA-256 equals the pack’s. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. The card's stated scope limit. A cross-family machine check (2026-10-30, item 39) re-read the record’s sources and found full support. No human editorial review. Status and meaning Editorial status. Not reviewed by an editor . An exploratory record that no editor has accepted. This status does not say whether a person has read any linked passage. Each evidence record above shows how it was checked.
Relation. A report states the result of an evaluation it ran or commissioned, in its own words, with what was measured, by what kind of method, and the caveats the report itself gives (schema 2.22). The record is the report’s statement, not a verified finding: it says nothing about another system, a later version or real-world risk, and a null or limiting result is not evidence that no effect exists.
Date. As stated in the card expression last updated 20 August 2025. The expression checked is the PDF served at data.x.ai/2025-08-20-grok-4-model-card.pdf on 2026-10-28, whose only date is that stated last update; the card does not say what any update changed, or when the card was first published.
Record ID ab-ev-grok4-biology-result