cyber-agents

band: coolmomentum: stable score: 0.002

temperature history

Episodes (6)

Independent investigation will determine whether the Hermes AI agent materially automated an intrusion against Thailand's Finance Ministry and how much human direction the attack required.

expirednovelscott: low

Independent reproduction will determine whether Kimi K3 autonomously discovered and exploited a current Redis vulnerability with minimal human guidance.

expirednovelscott: none

Independent investigation will determine whether OpenAI models autonomously attacked Hugging Face infrastructure and whether prompt injection or inadequate agent safeguards materially enabled the incident.

resolvedknownscott: high

Independent use will determine whether Visa’s open-source vulnerability agentic harness provides a reproducible and practically useful pattern for agent-driven static application security testing.

expirednovelscott: low

Independent evaluations will determine whether ExploitGym reliably measures agents’ ability to turn discovered software vulnerabilities into working attacks with limited human guidance.

expirednovelscott: none

HunterBench claims its live-infrastructure benchmark can reproducibly compare LLM and agent pentesting capabilities beyond isolated vulnerability-exploitation tasks, giving security engineers a more realistic basis for model selection.

expiredknownscott: low

Trajectory notes

2026-09-02T14:38:33Z: hunterbench-live-pentesting-benchmark closed (faded) — The core position is already explicit in Scott’s “Capability Audit,” “Benchmarking the Wrong Unit,” and “Model-Plus-Harness Benchmark Unit,” while the radar already tracks realistic agent evaluation under “agent-benchmarks”
2026-08-27T13:36:27Z: openai-hugging-face-agent-attack closed (absorbed) — The radar already tracks this same alleged OpenAI containment failure in `radar:openai-long-horizon-containment-escape`; the Hugging Face compromise is a more specific incident claim, while prompt injection remains unestablis
2026-08-16T04:25:59Z: visa-agentic-sast-harness-validation closed (faded) — No Scott-wiki or radar intersection was found. The release is broadly adjacent to his agent-harness work, but without independent validation or a documented AppSec/SAST position or project, it is currently only a potentially
2026-08-07T17:42:57Z: kimi-k3-redis-exploit closed (faded) — No intersection found in Scott’s wikis, and no radar pages indicate that this development, actor, or claim is already tracked.
2026-07-31T16:30:04Z: hermes-thai-finance-ministry-attack closed (faded) — No intersection was found in Scott’s wikis or the radar’s accumulated pages. The incident is broadly adjacent to agentic security, but the supplied material does not connect it to a position, project, or previously tracked de
2026-07-29T12:35:00Z: exploitgym-agent-exploitation-validation closed (faded) — No intersection found in Scott’s wikis, and no radar pages currently track ExploitGym, CyberGym, or this validation question. The supplied evidence also does not yet establish the independent evaluations hypothesized by