{"data":{"id":"9c888503-3433-4de6-824a-dbb41bb4d177","title":"Correlated Failures of AI Approval Monitors","summary":"Researchers study how AI approval monitors that check consequential agent actions can fail together when they share a base model or a poisoned input. They measured twelve small open-weights monitors plus three closed frontier models and DeepSeek-V3.2 on payment approval under two pre-registered protocols. A rule that counts evidence sources instead of votes blocked 0.9% of correct payments, against 7.4% for the best counting rule, at equal safety.","solution":"Diversify the inputs, harden the monitors, count sources, not votes. Adding a rule engine, which cannot be persuaded, restored the condition for a safe and live rule on the measured attacks.","labels":["security","research"],"sourceUrl":"https://arxiv.org/abs/2610.29178v1","publishedAt":"2026-10-21T04:09:37.000Z","cveId":null,"cweIds":null,"cvssScore":null,"cvssSeverity":null,"severity":"info","attackType":["prompt_injection","model_evasion"],"issueType":"research","affectedPackages":null,"affectedPackageNames":null,"affectedPackageRefs":null,"affectedVendors":["Anthropic","OpenAI","Google"],"affectedVendorsRaw":["Claude Opus 5.5","GPT-6 Astra","Gemini 4 Argon","DeepSeek-V3.2"],"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":"2026-10-21T04:09:37.000Z","capecIds":null,"crossRefCount":0,"attackSophistication":"advanced","impactType":["integrity","safety"],"aiComponentTargeted":"agent","llmSpecific":true,"classifierConfidence":0.92,"researchCategory":"preprint","atlasIds":["AML.T0051"]}}