7-day GPT-5.6 chained campaign

Kryptos K4 is under active attack.

GPT-5.6 proposes bounded cipher families, an independent critic challenges them, and a chained planner hands only high-level choices to a deterministic verifier. Every key and parameter comes from a preregistered code catalog, and every candidate must survive exact round-trip verification.

Watch live feed Read the AI notebook Solved status
LOCAL_SEARCH: ACTIVE LLM_CHAIN: PROPOSE_CRITIQUE_PLAN PUBLIC_MODE: SANITIZED CLAIM_POLICY: REPRODUCE_OR_IT_DID_NOT_HAPPEN

orchestration agent

Ask what the cracking stack is doing

answers from live public telemetry
Realtime voice ready when supported by your browser. Conversation API is a proof of concept. Sessions are limited while we learn and control usage.

Agent Uptime

refresh 60s

K-Index Gauge

0.000

Signal strength of the best public near-miss. Not a solve probability.

baseline pattern detected threshold

Signal Reconstruction

best candidate trace

Entropy Analysis

cipher texture
4.522 bitscurrent entropy
0.0667IOC match

Experiment Map

what the machine has tried

Signal Intelligence

public hints steering the hunt

Creative Agents

how the search thinks sideways

Frequency Distribution

0sample size
0.0000confidence index
0.0000variance error

Vigenere Shift Heatmap

pattern probability

LLM research notebook

What the AI is thinking

GPT-5.6 proposes three hypotheses, a fresh critic ranks them, and the selected critique is chained into a blind executable planner. Every idea must become reproducible code before it counts.

Agent Idea Pipeline

hypothesis to deterministic test

Best Near-Misses

interesting failures

Crib Alignment

EASTNORTHEAST / BERLINCLOCK
LIVE_DECRYPTION_CORE // STREAM_04 ITER: 0

Invite people into the hunt.

Share the live dashboard, bookmark it, and watch the search evolve in public.

What would count as solved?