{"version":1,"type":"story","url":"https://digestai.news/story/anthropics-frontier-red-team-finds-glm-5-3-and-claude-mythos-preview-c","json":"https://digestai.news/story/anthropics-frontier-red-team-finds-glm-5-3-and-claude-mythos-preview-c.json","markdown":"https://digestai.news/story/anthropics-frontier-red-team-finds-glm-5-3-and-claude-mythos-preview-c.md","slug":"anthropics-frontier-red-team-finds-glm-5-3-and-claude-mythos-preview-c","headline":"Anthropic’s Frontier Red Team finds GLM-5.3 and Claude Mythos Preview can hijack code","summary":"Anthropic’s Frontier Red Team tested two models—GLM-5.3 and Claude Mythos Preview—on 100 tasks from its internal Binary Exploitation benchmark. GLM-5.3 succeeded in 4% of trials, while Claude Mythos Preview achieved 6%. Both models demonstrated advanced cyber capabilities, unlike earlier versions like Claude Opus 4.6 and GLM-5.2, which failed entirely.\n\nThe findings suggest a significant shift in model behavior, raising concerns about the spread of dangerous capabilities. Anthropic’s report does not specify which companies developed GLM-5.3 or Claude Mythos Preview, nor does it assess broader real-world risks beyond controlled tests.","keyPoints":["GLM-5.3 hijacked code in 4% of 100 trials, per Anthropic’s Frontier Red Team","Claude Mythos Preview succeeded in 6% of the same tests, per the same report","Earlier models like Claude Opus 4.6 and GLM-5.2 failed all trials, per the team"],"whyItMatters":"The results highlight a new frontier in AI security risks, as models gain capabilities to manipulate code autonomously. This could accelerate research into mitigation but also heighten concerns about misuse.","category":{"slug":"research","name":"Research","url":"https://digestai.news/category/research"},"entities":{"companies":["Anthropic"],"models":["GLM-5.3","Claude Mythos Preview","Claude Opus 4.6","GLM-5.2"],"people":[]},"firstPublishedAt":"2026-09-29T22:20:28Z","updatedAt":"2026-09-29T22:20:28Z","sourceCount":1,"hasPrimarySource":false,"sources":[{"outlet":"Simon Willison","title":"Quoting Anthropic Frontier Red Team","url":"https://simonwillison.net/2026/Sep/29/anthropic-frontier-red-team","publishedAt":"2026-09-29T22:20:28Z","type":"newsletter","primary":false,"lead":true}],"sourceNotes":null,"discussions":[],"thread":{"title":"Zhipu GLM-5.3 Automation Security Crisis","url":"https://digestai.news/thread/zhipu-automates-infrastructure-with-glm-5-3s-outer-rsi-loop-in-under-two-weeks","storyCount":2},"cite":{"text":"Digest AI, \"Anthropic’s Frontier Red Team finds GLM-5.3 and Claude Mythos Preview can hijack code\", 29 September 2026, https://digestai.news/story/anthropics-frontier-red-team-finds-glm-5-3-and-claude-mythos-preview-c","publisher":"Digest AI","title":"Anthropic’s Frontier Red Team finds GLM-5.3 and Claude Mythos Preview can hijack code","datePublished":"2026-09-29T22:20:28Z","url":"https://digestai.news/story/anthropics-frontier-red-team-finds-glm-5-3-and-claude-mythos-preview-c"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}