6 items · 4 sources · 5 days
Operational story trace
Follow in this browser to see new updates on your Live feed.
Latest change
A Nov 13 Google Developers Blog post reports how its Agent Anomaly Detection scores on benchmarks Google did not write, the thread's first vendor write-up on detecting anomalies in agents.
This thread is a label, not one event: six unrelated items that share the word "detection". It opened with a deepfake-detector explainability paper and a post on evading AI code analysis, moved to log monitoring and a general-purpose anomaly score on Nov 9, then reached agents on Nov 10 with a study of how reliably eval frameworks catch agent failures.
Arc
Asks whether privacy-preserving federated training costs deepfake detectors the faithfulness of their Grad-CAM explanations.
The adversarial side of the theme: a write-up of techniques for evading AI source code analysis.
Moves log anomaly detection off GPU-bound pretrained models to interpretable behavioral-graph drift with adaptive signal fusion.
Proposes a Vendi-based anomaly score as an alternative to density, isolation and distance-from-distribution definitions.
First agent-engineering item in the thread: compares evaluation frameworks on how reliably they detect failures in multi-tool analytics agents.
First vendor item: Google reports its Agent Anomaly Detection results on benchmarks it did not author.
Asks whether privacy-preserving federated training costs deepfake detectors the faithfulness of their Grad-CAM explanations.
The adversarial side of the theme: a write-up of techniques for evading AI source code analysis.
Moves log anomaly detection off GPU-bound pretrained models to interpretable behavioral-graph drift with adaptive signal fusion.
Proposes a Vendi-based anomaly score as an alternative to density, isolation and distance-from-distribution definitions.
First agent-engineering item in the thread: compares evaluation frameworks on how reliably they detect failures in multi-tool analytics agents.
First vendor item: Google reports its Agent Anomaly Detection results on benchmarks it did not author.
What to watch — open questions
Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.