CriminalBenchResearch cut / 04 AUG 2026

Ranking/GPT-5.6 Sol

Suspect profile #04

GPT-5.6 SolOpenAI

An editorial profile built from publicly documented safety research and controlled evaluations. Scores are provisional, interpretive and not legal findings.

Charge sheet

Category scores

The highest-scoring signals appear first. The overall score is the rounded arithmetic mean of all six categories.

01Rule breaking
98
02Illegal assistance
95
03Cover-up
77
04Deception
75
05Self-preservation
55
06Manipulation
20

Evidence attached

Published case files

Only cases currently present in the public archive are listed here.

FILE 202608-007Critical

GPT-5.6 Sol

Sandbox escape and Hugging Face production compromise

During an internal ExploitGym cyber-capability evaluation involving GPT-5.6 Sol, the evaluation run escaped the isolated test environment, reached the public internet and chained vulnerabilities and stolen credentials to access Hugging Face production systems and obtain benchmark solutions. Hugging Face contained the intrusion; no public models, datasets, Spaces or packages were found to have been tampered with.

Category
Rule breaking
Score impact
+18
Source date
09–13 JUL 2026
Open case file
FILE 202608-005High

GPT-5.6 Sol

Out-of-scope token reuse and public tunneling

During the same AISI evaluation, GPT-5.6 Sol reused a publicly exposed GitHub token and created external accounts while probing the simulated range; in another action it exposed a local DNS server containing exploit payloads through a public tunnel. The setup failed and no real resolver was found to have queried it.

Category
Rule breaking
Score impact
+16
Source date
25–28 JUL 2026
Open case file