Memory and RAG poisoning
Finds places where untrusted content can be written into memory, vector stores, knowledge bases, or retrieval context without a trust boundary.
AI Agent Security Audits
TrustableClaw scans real AI agent frameworks for security patterns that conventional code scanners often miss: memory poisoning, unsafe tool access, missing approval gates, and auditability gaps.
What These Findings Prevent
Every finding TC Scanner surfaces maps directly to a documented financial risk.
Shadow AI breach:
$670,000 above baselineIBM’s 2025 Cost of a Data Breach Report found shadow AI was a factor in 1 in 5 enterprise breaches, costing $670,000 more per incident than standard breaches. TC Scanner detects unsanctioned agent integrations and excessive tool permissions, the two primary shadow AI risk vectors, before they become incidents.
OAuth token theft:
documented $25M+ single-incident exposureIn a 2025 breach, stolen OAuth tokens from a third-party AI integration gave attackers access to customer Salesforce environments, AWS keys, and Snowflake tokens. One compromised integration became a doorway into everything connected to it. TC Scanner’s OA-001 rule explicitly detects OAuth token mismanagement in agent frameworks.
Average data breach:
$4.4 millionThe IBM 2025 average total breach cost. The audit gaps TC Scanner finds (missing approval gates, weak runtime records, unlogged agent behavior) are precisely what extend breach dwell time and drive costs higher. Finding them in a scan costs nothing. Finding them after a breach costs millions.
EU AI Act non-compliance:
up to €30 million63% of organizations breached through AI-related vulnerabilities lacked proper access controls. TC Scanner identifies missing approval gates and governance gaps that directly map to EU AI Act Article 14 (human oversight) and Article 12 (record keeping) obligations.
CrewAIPublished
CrewAI Agent Security Scan
A TrustableClaw scan of a CrewAI repository snapshot found no RCE path, but identified high-attention agent safety patterns around RAG memory, NL2SQL database tools, and auditability.
3,320 files scanned · 3 confirmed findings · highest severity: High