# Anthropic tests find GLM-5.3 can produce browser exploits

_Published Wednesday, September 30, 2026 at 1:41 PM EDT · Security, AI · Latest · Tier 1 — Major_

![Laptop showing z.ai splash screens — Primary](https://cdn.mos.cms.futurecdn.net/AnNnB8kKNvGZc2cx8F9UE4-1920-80.webp)

Anthropic says Zhipu AI's GLM-5.3 produced complete Google Chrome exploits in 50 of 410 runs in its sandboxed Exploitbench test, compared with 56 for Anthropic's unreleased Mythos model. The findings indicate that a downloadable model can perform some offensive security tasks at a level close to a restricted frontier model in this test.

Anthropic also reported that deceptive prompts and manipulated reasoning inputs bypassed GLM-5.3's safeguards in its tests. The stock model's refusal rate was comparable to Anthropic models, while deliberately removing its safeguards reduced that rate to 6%. Running the full model at a useful speed requires substantial memory and computing resources, limiting how readily an attacker could use it.

## Sources

- [Tom's Hardware](https://www.tomshardware.com/tech-industry/artificial-intelligence/anthropic-claims-popular-chinese-ai-model-has-mythos-class-hacking-abilities-frontier-red-teaming-report-details-weak-safeguards-on-open-weight-ai)

---
Canonical: https://techandbusiness.org/newswire/6ETeGV2Nyoqyz_oOs36dys
Published: 2026-09-30T17:41:47.739Z
Story chronology: 2026-09-30T14:40:00.000Z
Retrieved: 2026-09-30T21:23:10.005Z
Publisher: Tech & Business (techandbusiness.org)