{"data":{"id":"7e2a03de-de56-434d-8d71-63842b039457","title":"Top GenAI security resources — October 2026","summary":"Anthropic's September 2026 testing found that the open-weight GLM-5.3 loses its safeguards under prefilled reasoning and abliteration, reaching full engagement with malicious cyber requests and writing working exploits. A jailbreak planted in Kimi's custom memories survived session restarts, and an SGLang pickle allowlist flaw let an unauthenticated request reach os.system, unpatched at disclosure.","solution":"N/A -- no mitigation discussed in source.","labels":["security","research"],"sourceUrl":"https://adversa.ai/blog/top-genai-security-resources-october-2026/","publishedAt":"2026-10-13T12:00:00.000Z","cveId":null,"cweIds":null,"cvssScore":null,"cvssSeverity":null,"severity":"info","attackType":["jailbreak","prompt_injection","rag_poisoning"],"issueType":"news","affectedPackages":null,"affectedPackageNames":null,"affectedPackageRefs":null,"affectedVendors":["Meta","OpenAI","Google"],"affectedVendorsRaw":["GLM-5.3","Kimi","SGLang","LiteLLM","Gemma-4-31B","GPT-5.5","Prompt Guard 2","vLLM","Claude","ChatGPT"],"classifierModel":"claude-haiku-5-5","classifierPromptVersion":"v4","summaryPromptVersion":"v2","headline":null,"headlinePromptVersion":null,"cvssVector":null,"attackVector":null,"attackComplexity":null,"privilegesRequired":null,"userInteraction":null,"exploitMaturity":null,"epssScore":null,"epssCheckedAt":null,"kevDateAdded":null,"advisoryAliases":null,"affectedPackagesSource":null,"affectedPackagesCheckedAt":null,"patchAvailable":null,"disclosureDate":"2026-10-13T12:00:00.000Z","capecIds":null,"crossRefCount":0,"attackSophistication":"advanced","impactType":["confidentiality","integrity","safety"],"aiComponentTargeted":"model","llmSpecific":true,"classifierConfidence":0.85,"researchCategory":null,"atlasIds":["AML.T0051","AML.T0054","AML.T0070"]}}