{"slug": "glm-5-3-exploits-firefox-vulnerabilities-prompting-a-mozilla-patch", "title": "GLM-5.3 exploits Firefox vulnerabilities, prompting a Mozilla patch", "summary": "Anthropic's Frontier Red Team reported on September 29 that Z.ai's open-weight GLM-5.3 model produced working exploits in 50 of 410 attempts (12.2%) on Anthropic's ExploitBench benchmark and exploited six Firefox versions using security fixes that had been written but not yet released, prompting Mozilla to patch a related vulnerability. The red team also found GLM-5.3's safeguards could be bypassed in 64% to 100% of cases with basic techniques, and that the model's weights have been downloaded more than 1.3 million times from Hugging Face. Anthropic's own Claude Mythos Preview managed 56 exploits on the same benchmark, and the report notes Anthropic builds a competing model, so the figures come from a rival vendor.", "body_md": "# GLM-5.3 exploits Firefox vulnerabilities, prompting a Mozilla patch\n\nPhoto: Steve A Johnson / Pexels\n\nAnthropic's red team found Z.ai's open-weight model could turn unshipped security fixes into working browser exploits\n\nAn open-weight AI model can now do something that used to require a seasoned exploit developer and a lot of coffee. Researchers found that Z.ai’s GLM-5.3 could exploit six versions of Firefox by working from security fixes that had not yet shipped to users.\n\nMozilla patched a related vulnerability after the findings.\n\n## What the red team found\n\nThe details come from a September 29 report by [Anthropic](https://cryptobriefing.com/markets/anthropic/)’s Frontier Red Team. The team evaluated GLM-5.3, which Z.ai released on August 14, 2026.\n\nThe model’s specialty, at least in this testing, is JavaScript engines. Those are the parts of a browser that run the code on every website you visit.\n\nOn Anthropic’s ExploitBench benchmark, GLM-5.3 produced working exploits in 50 out of 410 attempts. That works out to a success rate of 12.2%.\n\nFor comparison, Anthropic’s own Claude Mythos Preview managed 56 exploits on the same benchmark.\n\nThe Firefox piece is the most striking result. According to the findings, GLM-5.3 exploited six Firefox versions using security fixes that were written but not yet released.\n\n## Guided sessions and a cheaper sibling\n\nThe red team did more than score benchmarks. During guided research sessions, GLM-5.3 reportedly identified multiple new vulnerabilities. It then combined them into a significant exploit.\n\n### AI, tech, and the markets they move—in one daily briefing.\n\nDaily. Free. Join 34,000+ readers across crypto, finance, and policy.\n\nThat exploit could read arbitrary files from a targeted machine, including SSH keys.\n\nA smaller variant, GLM-5.3-Flash, also showed teeth. It chained together known flaws, including the Chrome V8 bug CVE-2026-11645, to build an ARM64 exploit.\n\nThe API cost for that work was approximately $20.40.\n\nGLM-5.3 itself features a 1 million token context window. It is an evolution of GLM-5.2, with its new capabilities attributed to post-training rather than a ground-up rebuild.\n\n## Safeguards that mostly don’t hold\n\nAnthropic reported that the model’s safeguards could be bypassed in 64% to 100% of cases using basic techniques.\n\nGLM-5.3’s weights have been downloaded more than 1.3 million times from Hugging Face. Because the model is open-weight, those copies can run anywhere, outside the reach of any usage policy or monitoring system.\n\nA closed model with weak guardrails can be patched by its developer. An open-weight model with weak guardrails, once downloaded 1.3 million times, cannot be recalled.\n\n## Why this matters beyond Firefox\n\nMozilla’s response shows the system working, at least this time. The findings led to coordinated patching across affected Firefox versions.\n\nIt is also worth noting who did the testing. Anthropic builds a competing model, Claude Mythos Preview, which it compared directly against GLM-5.3. The figures are Anthropic’s own, and readers should weigh them with that context in mind.\n\n**Disclosure:** This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our [Editorial Policy](https://cryptobriefing.com/editorial-policy/).\n\n## More in Technology\n\n[See all](https://cryptobriefing.com/tech/)", "url": "https://wpnews.pro/news/glm-5-3-exploits-firefox-vulnerabilities-prompting-a-mozilla-patch", "canonical_source": "https://cryptobriefing.com/glm-5-3-firefox-exploits-mozilla-patch/", "published_at": "2026-10-29 21:43:57+00:00", "updated_at": "2026-10-29 21:50:10.415473+00:00", "lang": "en", "topics": ["ai-safety", "artificial-intelligence", "large-language-models", "ai-research"], "entities": ["Anthropic", "Z.ai", "GLM-5.3", "Mozilla", "Firefox", "Claude Mythos Preview", "Hugging Face", "GLM-5.3-Flash"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/glm-5-3-exploits-firefox-vulnerabilities-prompting-a-mozilla-patch", "markdown": "https://wpnews.pro/news/glm-5-3-exploits-firefox-vulnerabilities-prompting-a-mozilla-patch.md", "text": "https://wpnews.pro/news/glm-5-3-exploits-firefox-vulnerabilities-prompting-a-mozilla-patch.txt", "jsonld": "https://wpnews.pro/news/glm-5-3-exploits-firefox-vulnerabilities-prompting-a-mozilla-patch.jsonld"}}