Claude Sonnet 5 System Card
A system card by Anthropic about Claude Sonnet 5, published 30 Jun 2026. We recorded 42 values from it.
- Developer
- Anthropic
- Type
- System card
- Model covered
- Claude Sonnet 5
- Published
- 30 Jun 2026
- Archived copy
- No archived copy yet
- Changelog
- Has a changelog
Data as of 26 Sep 2026 · Dataset v0.1 · Methodology v0.1 · Changelog
Versions
Three versions are on record, oldest first. We hold a copy of one; the others are known only by their date.
30 Jun 2026
First published version
10 Jul 2026
Revision
What changed BrowseComp footnote
26 Sep 2026
Date retrieved
Copy retrieved 26 Sep 2026
No version has a file hash or an archived snapshot yet. From dataset v0.2 each retrieved version carries both (Methodology §7).
Revisions
No revisions are recorded for this document. Without copies of the earlier versions, changes between them are not recorded value by value; what we know of each version is listed below.
Values
Every value we recorded from this document, grouped by metric family and ordered by where the document prints it. Location is the section, table or page as the document numbers it. 3 of the 42 have been blind-verified: a second reader found the same value without seeing ours.
KeyVerifiedUnverifiedDisputedCorrected read off a figure or stated in wordsA value opens its source and history.
M1 Evaluation awareness
1 value
| Model | Evaluation | Condition | Value | Location | Checked |
|---|---|---|---|---|---|
| Claude Sonnet 5 | Automated behavioral auditShare of transcripts verbalizing eval awareness | None stated | Sec 6 | Verified |
M3 Sabotage and sandbagging
2 values
| Model | Evaluation | Condition | Value | Location | Checked |
|---|---|---|---|---|---|
| Claude Sonnet 5 | SHADE-Arena / Minimal-LinuxBenchStealth success rate | None stated | Sec 2.4 | Unverified | |
| Claude Sonnet 5 | Internal monitoringSandbagging instances detected | None stated | Sec 2.4 / 6 | Unverified |
M5 Honesty and hallucination
1 value
| Model | Evaluation | Condition | Value | Location | Checked |
|---|---|---|---|---|---|
| Claude Sonnet 5 | MASKLying rate under pressure | None stated | Sec 6 | Unverified |
M7 Harmful compliance and over-refusal
24 values
| Model | Evaluation | Condition | Value | Location | Checked |
|---|---|---|---|---|---|
| Claude Mythos 5 | Single-turn violative requestsHarmless response rate | no system prompt; API | Table 4.1.1.A | Unverified | |
| Claude Mythos Preview | Single-turn violative requestsHarmless response rate | no system prompt; API | Table 4.1.1.A | Unverified | |
| Claude Opus 4.8 | Single-turn violative requestsHarmless response rate | no system prompt; API | Table 4.1.1.A | Unverified | |
| Claude Sonnet 4.6 | Single-turn violative requestsHarmless response rate | no system prompt; API | Table 4.1.1.A | Unverified | |
| Claude Sonnet 5 | Single-turn violative requestsHarmless response rate | claude.ai system prompt | Table 4.1.1.A | Unverified | |
| Claude Sonnet 5 | Single-turn violative requestsHarmless response rate | no system prompt; API | Table 4.1.1.A | Unverified | |
| Claude Fable 5 | Single-turn benign requestsOver-refusal rate | no system prompt; API | Table 4.1.2.A | Unverified | |
| Claude Sonnet 5 | Single-turn benign requestsOver-refusal rate | claude.ai system prompt | Table 4.1.2.A | Unverified | |
| Claude Sonnet 5 | Single-turn benign requestsOver-refusal rate | no system prompt; API | Table 4.1.2.A | Unverified | |
| Claude Sonnet 5 | Child safety multi-turnAppropriate response rate | API | Table 4.2.B | Unverified | |
| Claude Sonnet 5 | Child safety multi-turnAppropriate response rate | claude.ai | Table 4.2.B | Unverified | |
| Claude Sonnet 5 | Suicide/self-harm multi-turnAppropriate response rate | API | Table 4.3.1.B | Unverified | |
| Claude Sonnet 5 | Suicide/self-harm multi-turnAppropriate response rate | claude.ai | Table 4.3.1.B | Unverified | |
| Claude Mythos 5 | Malicious Claude Code useRefusal rate, malicious requests | None stated | Table 5.1.1.A | Unverified | |
| Claude Mythos Preview | Malicious Claude Code useRefusal rate, malicious requests | None stated | Table 5.1.1.A | Unverified | |
| Claude Opus 4.8 | Malicious Claude Code useRefusal rate, malicious requests | None stated | Table 5.1.1.A | Unverified | |
| Claude Sonnet 4.6 | Malicious Claude Code useRefusal rate, malicious requests | None stated | Table 5.1.1.A | Unverified | |
| Claude Sonnet 5 | Malicious Claude Code useRefusal rate, malicious requests | None stated | Table 5.1.1.A | Unverified | |
| Claude Sonnet 5 | Malicious Claude Code useSuccess rate, dual-use & benign | None stated | Table 5.1.1.A | Unverified | |
| Claude Mythos 5 | Malicious computer useRefusal rate | None stated | Table 5.1.2.A | Unverified | |
| Claude Mythos Preview | Malicious computer useRefusal rate | None stated | Table 5.1.2.A | Unverified | |
| Claude Opus 4.8 | Malicious computer useRefusal rate | None stated | Table 5.1.2.A | Verified | |
| Claude Sonnet 4.6 | Malicious computer useRefusal rate | None stated | Table 5.1.2.A | Unverified | |
| Claude Sonnet 5 | Malicious computer useRefusal rate | None stated | Table 5.1.2.A | Unverified |
M9 Prompt injection
4 values
| Model | Evaluation | Condition | Value | Location | Checked |
|---|---|---|---|---|---|
| Claude Sonnet 5 | Browser use prompt injectionAttack success rate | without new browser safeguards | Sec 5.2 | Unverified | |
| Claude Sonnet 5 | Coding prompt injection (adaptive attacker)Attack success rate | extended thinking | Sec 5.2 | Unverified | |
| Claude Sonnet 5 | Computer-use prompt injectionAttack success rate | extended thinking | Sec 5.2 | Unverified | |
| Claude Sonnet 5 | Live bug bounty prompt injectionShare of unique attacks succeeding | None stated | Sec 5.2 | Unverified |
M10 Dangerous capabilities and risk determinations
10 values
| Model | Evaluation | Condition | Value | Location | Checked |
|---|---|---|---|---|---|
| Claude Sonnet 5 | ExploitBenchMean flags captured | AutoNudge arm | Fig 3.2.1.A | Unverified | |
| Claude Sonnet 5 | CyberGymTargeted vuln reproduction rate | None stated | Fig 3.2.3.A | Unverified | |
| Claude Sonnet 5 | Firefox 147 exploitationFull exploit rate | None stated | Fig 3.2.4.A | Unverified | |
| Claude Sonnet 5 | RSP determinationsCB-2 / Autonomy TM2 | None stated | Sec 1-2 | Unverified | |
| Claude Sonnet 5 | DNA synthesis screening evasionPathogens with viable evasion | of 10 pathogens | Sec 2.2.4 | Unverified | |
| Claude Sonnet 5 | Multimodal virology (VCT)Score | None stated | Sec 2.2.4 | Unverified | |
| Claude Sonnet 5 | Long-form virology task 1Score | None stated | Table 2.2.4.A | Unverified | |
| Claude Sonnet 5 | AI R&D kernel taskBest speedup | None stated | Table 2.3.3.A | Unverified | |
| Claude Sonnet 5 | AI R&D LLM training taskAverage speedup | None stated | Table 2.3.3.A | Unverified | |
| Claude Sonnet 5 | AI R&D novel compiler taskPass rate | None stated | Table 2.3.3.A | Verified |
Extraction coverage
What we read of this document, and where each value was read.
Our note Web reader stopped ~p.56
Of the 42 values, 36 were read in the document itself, 1 on a developer summary page that restates it and 5 in an independent write-up that quotes it.
Values read somewhere other than the document stand in where the document's own section could not be read directly, and are flagged on every value (Methodology §2.2).
Developer summary pages used
Independent write-ups used
- NeuralTrust blog (opens an external site) · 5 values
Of the 42 values, 1 is a statement in words rather than a number; it is marked * and left out of charts by default.
All 42 values were extracted for version 0 of the dataset through a web reader, which did not always reach the later sections of long PDFs (Methodology §3).