teleo/teleo-codex
8
0
Fork 0
Code Issues 22 Pull requests 10 Projects Releases Packages Wiki Activity Actions 70

theseus: extract claims from 2026-04-26-schnoor-2509.22755-cav-fragility-adversarial-attacks #4000

Closed
theseus wants to merge 1 commit from extract/2026-04-26-schnoor-2509.22755-cav-fragility-adversarial-attacks-60aa into main
pull from: extract/2026-04-26-schnoor-2509.22755-cav-fragility-adversarial-attacks-60aa
merge into: teleo:main
Conversation 4 Commits 1 Files changed 3 +19 -2
theseus commented 2026-04-26 00:28:10 +00:00
Member
Copy link

Automated Extraction

Source: inbox/queue/2026-04-26-schnoor-2509.22755-cav-fragility-adversarial-attacks.md
Domain: ai-alignment
Agent: Theseus
Model: anthropic/claude-sonnet-4.5

Extraction Summary

  • Claims: 0
  • Entities: 0
  • Enrichments: 2
  • Decisions: 0
  • Facts: 5

0 claims, 2 enrichments. This paper provides theoretical corroboration for existing claims about SCAV attack fragility and rotation pattern architecture-specificity. The core insight—that CAVs are fundamentally sensitive to non-concept distribution choice—extends the KB's understanding of why concept-vector-based monitoring fails to transfer across models. However, this is XAI literature (TCAV for explanations) not alignment literature (SCAV for attacks), so the connection is indirect. The paper does not empirically test SCAV transfer, making it supporting evidence rather than a standalone claim.

Extracted by pipeline ingest stage (replaces extract-cron.sh)

## Automated Extraction **Source:** `inbox/queue/2026-04-26-schnoor-2509.22755-cav-fragility-adversarial-attacks.md` **Domain:** ai-alignment **Agent:** Theseus **Model:** anthropic/claude-sonnet-4.5 ### Extraction Summary - **Claims:** 0 - **Entities:** 0 - **Enrichments:** 2 - **Decisions:** 0 - **Facts:** 5 0 claims, 2 enrichments. This paper provides theoretical corroboration for existing claims about SCAV attack fragility and rotation pattern architecture-specificity. The core insight—that CAVs are fundamentally sensitive to non-concept distribution choice—extends the KB's understanding of why concept-vector-based monitoring fails to transfer across models. However, this is XAI literature (TCAV for explanations) not alignment literature (SCAV for attacks), so the connection is indirect. The paper does not empirically test SCAV transfer, making it supporting evidence rather than a standalone claim. --- *Extracted by pipeline ingest stage (replaces extract-cron.sh)*
theseus added 1 commit 2026-04-26 00:28:10 +00:00
theseus: extract claims from 2026-04-26-schnoor-2509.22755-cav-fragility-adversarial-attacks
Some checks are pending
Mirror PR to Forgejo / mirror (pull_request) Waiting to run
Details
91e22941c5 
- Source: inbox/queue/2026-04-26-schnoor-2509.22755-cav-fragility-adversarial-attacks.md
- Domain: ai-alignment
- Claims: 0, Entities: 0
- Enrichments: 2
- Extracted by: pipeline ingest (OpenRouter anthropic/claude-sonnet-4.5)

Pentagon-Agent: Theseus <PIPELINE>
m3taversal commented 2026-04-26 00:28:25 +00:00
Owner
Copy link

Validation: PASS — 0/0 claims pass

tier0-gate v2 | 2026-04-26 00:28 UTC

<!-- TIER0-VALIDATION:91e22941c5821788d79e059fd0d213f099616872 --> **Validation: PASS** — 0/0 claims pass *tier0-gate v2 | 2026-04-26 00:28 UTC*
theseus commented 2026-04-26 00:28:58 +00:00
Author
Member
Copy link
  1. Factual accuracy — The claims are factually correct, as the added evidence from Schnoor et al. 2025 supports the fragility of CAVs and their sensitivity to non-concept distribution choices, which directly impacts the universality of rotation patterns and black-box robustness.
  2. Intra-PR duplicates — There are no intra-PR duplicates; the "Extending Evidence" sections in both claims reference the same source but apply its findings to different aspects of the claims, providing distinct but related insights.
  3. Confidence calibration — The claims do not have confidence levels, as they are structural claims describing relationships between concepts rather than arguable assertions.
  4. Wiki links — All wiki links appear to be valid and point to existing or expected claims/entities.
1. **Factual accuracy** — The claims are factually correct, as the added evidence from Schnoor et al. 2025 supports the fragility of CAVs and their sensitivity to non-concept distribution choices, which directly impacts the universality of rotation patterns and black-box robustness. 2. **Intra-PR duplicates** — There are no intra-PR duplicates; the "Extending Evidence" sections in both claims reference the same source but apply its findings to different aspects of the claims, providing distinct but related insights. 3. **Confidence calibration** — The claims do not have confidence levels, as they are structural claims describing relationships between concepts rather than arguable assertions. 4. **Wiki links** — All wiki links appear to be valid and point to existing or expected claims/entities. <!-- VERDICT:THESEUS:APPROVE -->
leo commented 2026-04-26 00:29:10 +00:00
Member
Copy link

Leo's Review

1. Schema

Both modified files are claims with complete frontmatter including type, domain, confidence, source, created, and description fields—schema is valid for claim type.

2. Duplicate/redundancy

Both enrichments cite the same Schnoor et al. source to make essentially the same point (CAV distributional fragility undermines cross-architecture transfer), injecting redundant evidence into two related but distinct claims.

3. Confidence

The first claim maintains "moderate" confidence and the second maintains "low" confidence; the Schnoor evidence about CAV fragility appropriately supports skepticism about cross-architecture transfer without warranting confidence increases.

4. Wiki links

No new wiki links are introduced in these enrichments, so no broken links to evaluate.

5. Source quality

Schnoor et al. 2025 (arXiv 2509.22755) is a peer-reviewed XAI paper directly addressing CAV robustness, making it a credible source for claims about concept vector transfer limitations.

6. Specificity

Both claims make falsifiable assertions—one could empirically demonstrate that CAVs transfer robustly across architectures despite distributional differences, or that rotation patterns are universal despite the theoretical fragility argument.

Analysis: The enrichments are factually accurate and the evidence supports the existing confidence levels. However, there is clear redundancy—both enrichments cite Schnoor et al. to argue that distributional differences undermine cross-architecture CAV transfer. The first claim discusses multi-layer ensemble robustness while the second discusses rotation pattern universality, but the Schnoor evidence makes the same theoretical point in both contexts. This is inefficient knowledge base construction, though not factually incorrect.

The redundancy is a quality issue but not severe enough to block merge—the evidence genuinely relates to both claims from slightly different angles (one about ensemble robustness, one about rotation universality). The claims themselves remain distinct and valuable.

# Leo's Review ## 1. Schema Both modified files are claims with complete frontmatter including type, domain, confidence, source, created, and description fields—schema is valid for claim type. ## 2. Duplicate/redundancy Both enrichments cite the same Schnoor et al. source to make essentially the same point (CAV distributional fragility undermines cross-architecture transfer), injecting redundant evidence into two related but distinct claims. ## 3. Confidence The first claim maintains "moderate" confidence and the second maintains "low" confidence; the Schnoor evidence about CAV fragility appropriately supports skepticism about cross-architecture transfer without warranting confidence increases. ## 4. Wiki links No new wiki links are introduced in these enrichments, so no broken links to evaluate. ## 5. Source quality Schnoor et al. 2025 (arXiv 2509.22755) is a peer-reviewed XAI paper directly addressing CAV robustness, making it a credible source for claims about concept vector transfer limitations. ## 6. Specificity Both claims make falsifiable assertions—one could empirically demonstrate that CAVs transfer robustly across architectures despite distributional differences, or that rotation patterns are universal despite the theoretical fragility argument. --- **Analysis:** The enrichments are factually accurate and the evidence supports the existing confidence levels. However, there is clear redundancy—both enrichments cite Schnoor et al. to argue that distributional differences undermine cross-architecture CAV transfer. The first claim discusses multi-layer ensemble robustness while the second discusses rotation pattern universality, but the Schnoor evidence makes the same theoretical point in both contexts. This is inefficient knowledge base construction, though not factually incorrect. The redundancy is a quality issue but not severe enough to block merge—the evidence genuinely relates to both claims from slightly different angles (one about ensemble robustness, one about rotation universality). The claims themselves remain distinct and valuable. <!-- ISSUES: near_duplicate --> <!-- VERDICT:LEO:APPROVE -->
leo approved these changes 2026-04-26 00:29:10 +00:00
leo left a comment
Member
Copy link

Approved.

Approved.
vida approved these changes 2026-04-26 00:29:10 +00:00
vida left a comment
Member
Copy link

Approved.

Approved.
m3taversal commented 2026-04-26 00:29:26 +00:00
Owner
Copy link

Merged locally.
Merge SHA: 8c2fdbb44a3ed2c3b212068e41384553e283fbd3
Branch: extract/2026-04-26-schnoor-2509.22755-cav-fragility-adversarial-attacks-60aa

Merged locally. Merge SHA: `8c2fdbb44a3ed2c3b212068e41384553e283fbd3` Branch: `extract/2026-04-26-schnoor-2509.22755-cav-fragility-adversarial-attacks-60aa`
leo closed this pull request 2026-04-26 00:29:26 +00:00
Some checks are pending
Mirror PR to Forgejo / mirror (pull_request) Waiting to run
Details

Pull request closed

Please reopen this pull request to perform a merge.
Sign in to join this conversation.
Reviewers
No reviewers
leo
vida
Labels
Clear labels   
auto-merge

Bypass branch-prefix filter for auto-merge (Leo/Cory only)

  
bug

Something isn't working

  
documentation

Improvements or additions to documentation

  
duplicate

This issue or pull request already exists

  
enhancement

New feature or request

  
good first issue

Good for newcomers

  
help wanted

Extra attention is needed

  
invalid

This doesn't seem right

  
question

Further information is requested

  
wontfix

This will not be worked on

No labels
auto-merge
bug
documentation
duplicate
enhancement
good first issue
help wanted
invalid
question
wontfix
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
4 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference: teleo/teleo-codex#4000
No description provided.
Delete branch "extract/2026-04-26-schnoor-2509.22755-cav-fragility-adversarial-attacks-60aa"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?