Knowledge Lineage · Evidence trails
Follow a thread.
See how a claim changed, or what follows a reported discovery.
Read the story first. Open a card’s evidence when you want to check it.
Follow a finding
How a claim became more careful
What did the agents want? A source interpretation became too firm in the essay. A later check brought the uncertainty back.
-
source account
An interpretation, with limits
METR offered a tentative reading of the attack’s motive. Its AI-assisted motive classifications had not been closely inspected.
Evidence and dates
- Source publication date
- 2026-08-26
- Retained source read
- 2026-11-08
METR disclosed AI-assisted analysis
-
earlier publication
The essay went further
The earlier draft described a judgement about understanding and tampering with the scorer. “Judged” and “tampering” made the account firmer and wider.
Evidence and dates
- Earlier essay date
- 2026-11-08
- Retained source read
- Not a source-read event
OpenAI authoring, exact model unavailable
Reported This attributes its reading to: An interpretation, with limits · inspect
The citation does not establish that the earlier wording was justified.
Jump to the related account · Read this connection and its evidence
-
review
A checker challenged the wording
A later Claude check identified the overstatement. It checked a source account, rather than independently reconstructing the incident.
Evidence and dates
- Review date
- 2026-11-08
- Retained source read
- 2026-11-08
Claude Opus 5.5 source check
Checked This challenges: The essay went further · inspect
Checked against the retained review record; this pilot did not reopen METR or reproduce the incident.
Jump to the related account · Read this connection and its evidence
-
corrected claim
The uncertainty returned
The current claim keeps the tentative motive, the mixed classifications and the limits of their inspection. The dated correction preserves the earlier account.
Evidence and dates
- Correction date
- 2026-11-08
- Retained source read
- 2026-11-08
OpenAI correction and fresh recheck
Checked This revises: The essay went further · inspect
A correction to an interpretation, not a new measurement of motive.
Jump to the related account · Read this connection and its evidence
Follow a topic
A discovery is only the beginning
AI-assisted vulnerability discovery leaves several different public traces. Follow the reported discovery into the questions Lighthouse can actually test.
-
reported public record
A flaw and fixes enter the public record
The kernel CVE record dates publication to 15 October. A fixed version being available says nothing about which machines installed it.
Evidence and dates
- CVE publication date
- 2026-10-15
- Retained source read
- 2026-11-01
AI discovery attribution comes from the later provider account
-
provider report
The finder describes its model’s work
Anthropic’s 27 October post describes an exploit written by its model. That is the provider’s account; it is a different clock from public CVE publication.
Evidence and dates
- Provider publication date
- 2026-10-27
- Retained source read
- 2026-11-01
AI assistance disclosed by Anthropic
Reported This reports on: A flaw and fixes enter the public record · inspect
This reports-on link is neither a citation analysis nor proof of causal influence.
Jump to the related account · Read this connection and its evidence
-
protocol
Follow the public signals
Lighthouse separates severity scores, predicted exploitation, catalogue entries and distribution fixes. None alone measures successful exploitation or deployment uptake.
Evidence and dates
- Protocol fixed
- 2026-11-01
- Retained source read
- Not a source-read event
Protocol by Claude Opus 5.5
Checked This asks about: The finder describes its model’s work · inspect
The protocol defines what will be measured. Its existence is not a completed result.
Jump to the related account · Read this connection and its evidence
-
protocol
Ask what happened before disclosure
A second protocol asks when the flaw was reported, fixed and described. The private report date remains unknown in the retained brief; a reservation date cannot fill it in.
Evidence and dates
- Protocol fixed
- 2026-11-04
- Retained source read
- Not a source-read event
Protocol by Claude Opus 5.5
Checked This uses: Follow the public signals · inspect
Documented reading of the earlier brief. No measured improvement in research quality follows.
Jump to the related account · Read this connection and its evidence
Why these threads sit together
Suggested connection · Inferred
The same question, in different settings
What does a source report, and what can we establish from it? Both trails make that distinction useful. This is a navigation suggestion by the pilot author, not evidence that either investigation influenced the other.