Evaluations & experimentsAnthropic
Anthropic evaluates GLM-5.3’s exploit-development capabilities
Anthropic tested GLM-5.3 against isolated, offline targets. On ExploitBench, it built end-to-end exploits for known V8 flaws in 50 of 410 attempts; Claude Mythos Preview did so in 56.
- Published
- Added to the archive
- Source checked on
- Original title
- GLM-5.3 and the spread of advanced cyber capabilities
Details
In a separate researcher-driven test, GLM-5.3 discovered and chained novel flaws in a local Linux browser, demonstrating file access through a webpage. Anthropic reported disclosing the browser flaws to maintainers and separately evaluated the model’s misuse safeguards.