One Word Opens the Gate: The Option-Channel Attack on Typed Decision Models as Agent Guardrails
Introduced the option-channel attack on typed decision models used as agent guardrails.
AI Storyline
3 items · 3 sources · 3 days
Operational story trace
Typed Decision
Follow in this browser to see new updates on your Live feed.
Current stateDeveloping
Latest change
A paper on routing a personal multilingual inbox compares typed decision models against a classical baseline and a generative LLM, arguing to measure before training.
Earlier contextThe story so far
Typed decision models return a probability over caller-defined options instead of generating text, and papers began placing them in agent systems as guardrails. The first paper showed an attack on that role; a security-triage study followed.
editor-curated · source-linked
Arc
Oct 8Oct 23 · now
RISK · Oct 8Option-channel attack on typed decision guardrails1 source · show source ▾
APPLICATION · Oct 21Typed, calibrated heads for perimeter triage1 source · show source ▾
VX-One: Typed, Calibrated Decision Heads for Real-Time Perimeter Triage - An empirical study of unknown-attack honesty, cascade gray zones, and calibration under drift
Applied typed, calibrated decision heads to real-time perimeter triage, contrasting them with signature engines and generative models.
NOW · Oct 23Measuring typed models against classical and generative baselines1 source · show source ▾
Measure Before You Train: Typed Decision Models vs. a Classical Baseline vs. a Generative LLM for Routing a Personal Multilingual Inbox
Compared typed decision models with a classical baseline and a generative LLM for inbox routing.
Day 1 Thursday, Oct 8, 2026
One Word Opens the Gate: The Option-Channel Attack on Typed Decision Models as Agent Guardrails
Introduced the option-channel attack on typed decision models used as agent guardrails.
Day 2 Wednesday, Oct 21, 2026
VX-One: Typed, Calibrated Decision Heads for Real-Time Perimeter Triage - An empirical study of unknown-attack honesty, cascade gray zones, and calibration under drift
Applied typed, calibrated decision heads to real-time perimeter triage, contrasting them with signature engines and generative models.
Day 3 Friday, Oct 23, 2026
Measure Before You Train: Typed Decision Models vs. a Classical Baseline vs. a Generative LLM for Routing a Personal Multilingual Inbox
Compared typed decision models with a classical baseline and a generative LLM for inbox routing.
What to watch — open questions
- Do any vendors or papers publish defenses against the option-channel attack?
- Do typed decision models hold their advantage over classical baselines on other routing tasks?
How this thread was built
editor wrote the arc · 3 beatswatcher 1 status change
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.