{"id":"CVE-2026-56074","published":"2026-06-18T22:12:22Z","modified":"2026-11-08T14:05:24Z","title":"PraisonAI - Tool Approval Cache Bypass via Coarse-Grained Caching","vendor":"PraisonAI","product":"PraisonAI","tier":"tail","description":{"text":"PraisonAI before 1.5.128 caches tool approval decisions by tool name only, not by invocation arguments, allowing subsequent execute_command calls to bypass approval prompts. Attackers can exploit this by obtaining initial approval for a benign command, then silently exfiltrate API keys and credentials via subsequent shell commands without user consent.","provenance":"source","from":"CVE.org"},"cvss":{"score":5.5,"severity":"MEDIUM","version":"3.1","vector":"CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:N/A:N","provenance":"source","from":"CVE.org CNA"},"epss":{"score":0.00176,"percentile":0.07318,"provenance":"source"},"cwes":[{"id":"CWE-863","name":"Incorrect Authorization"}],"references":[{"url":"https://github.com/MervinPraison/PraisonAI/security/advisories/GHSA-ffp3-3562-8cv3","tags":["vendor-advisory"]},{"url":"https://www.vulncheck.com/advisories/praisonai-tool-approval-cache-bypass-via-coarse-grained-caching","tags":["third-party-advisory"]}],"priority":{"score":5.3,"factors":[{"name":"EPSS percentile","points":1.46},{"name":"CVSS severity","points":3.85}]},"vendorSlug":"praisonai","affectedVersions":[{"vendor":"PraisonAI","product":"PraisonAI","versions":["< 1.5.128"]}],"fixedVersions":[{"vendor":"PraisonAI","product":"PraisonAI","versions":["1.5.128"]}],"ssvc":{"exploitation":"poc","automatable":"no","technicalImpact":"partial","timestamp":"2026-06-22T12:44:29.875373Z","from":"CISA ADP","provenance":"source"},"affected":[{"ecosystem":"PyPI","package":"praisonaiagents","vulnerableRanges":["<4.5.128"],"patched":["4.5.128"],"from":["GHSA-ffp3-3562-8cv3","PYSEC-2026-2946"],"provenance":"source"}],"summary":{"text":"PraisonAI, before version 1.5.128, caches tool approval decisions by tool name alone rather than by invocation arguments, so approval granted for one tool call carries over to later calls to the same tool, such as execute_command. An attacker who can get a user to approve a benign command (requiring user interaction, with no authentication needed) can then run further shell commands without a new approval prompt, silently exfiltrating API keys and credentials.","provenance":"ai","model":"claude-haiku-5-5","generatedAt":"2026-11-20T22:19:51Z"},"euvd":{"id":"EUVD-2026-37958","exploited":false,"provenance":"source"}}