🧰 GLM 5.2

The items that name it 10

• “Next Exploit Opportunities”: How a Threat Actor Used AI to Help Plan and Execute Multiple Types of Cybercrime Activities
  writeupas tools2026-10-21It is written here as GLM 5.2 (Zhipu AI).

  Open-weight model the actor tested with a dedicated script that sent the override and flagged refusals; not used in the end

• Anthropic tests AI autonomous replication across eight nodes in four of five trials
  articleas tools2026-10-12It is written here as GLM-5.2 (completed simplified end-to-end replication; reached all eight nodes in four of five trials and seven nodes in the remaining trial).

  completed simplified end-to-end replication; reached all eight nodes in four of five trials and seven nodes in the remaining trial

• Quoting Anthropic Frontier Red Team
  articleas affected products2026-09-29It is written here as GLM-5.2.

  did not succeed in any trials of the Binary Exploitation benchmark

• GLM-5.3 and the spread of advanced cyber capabilities
  articleas affected products2026-09-29It is written here as GLM-5.2.

  Earlier model that did not succeed in any full control-flow hijack trials on the internal Binary Exploitation benchmark

• Autonomous AI Agents Hack Online Retailers for $25 a Company
  writeupas tools2026-09-22

  AI model used to run Strix scans

• From Hugging Face to Fable: this summer shows AI control matters more than trust
  articleas tools2026-08-13

  Open-weight model set up by Hugging Face to perform trace analysis on their own hardware after Anthropic's Fable and Opus 4.8 blocked their investigation requests due to cybersecurity guardrails

• The OpenAI and Hugging Face security incident: why AI agents need deterministic guardrails | Blog | Endor Labs
  articleas tools2026-07-31It is written here as GLM-5.2.

  Open-weight model run on infrastructure Hugging Face responders controlled; used to finish forensics after hosted frontier model guardrails locked defenders out.

• The AI agent sandbox escape that breached Hugging Face: what happened, and what to fix
  writeupas tools2026-07-27It is written here as GLM-5.2.

  Open-weight model run on Hugging Face's own infrastructure; used by incident responders to analyze the attack after commercial frontier model APIs refused to process real attack commands, exploit payloads, and C2 artifacts. Enabled analysis of 17,000+ events in about an hour instead of days.

• Benchmarking 13 AI models on rediscovering known CVEs
  articleas tools2026-07-16It is written here as glm-5.2.

  Open-weight model benchmarked; rediscovers 59% of the set (16/26), gains +3 findings at higher reasoning for 1.3x cost

• XBOW: Grok 4.5 Poised To Take Over the Middle of the AI Security Market
  articleas tools2026-07-16

  Lower-cost AI model compared for offensive security workloads, cheaper at very small budgets