Plugin & Extension Security for AI: Complete Guide | QuizBy Eyal Doron / December 6, 2025 / 1 minute of reading Plugin & Extension Security for AI: Complete Guide | Quiz 1 / 7 1. According to the article – why is the misconception that plugins just read data not harmful dangerous? 1. Many plugins have write and execute capabilities and even read-only plugins can exfiltrate data 2. Reading data requires special permissions that are never granted 3. All plugins are strictly read-only by default 4. Data exfiltration is impossible through plugin architecture Correct! Why: Many plugins have write and execute capabilities and even read-only plugins can exfiltrate sensitive data – read access to the wrong data is still a breach. Context: The assumption that reading is harmless ignores both hidden capabilities and data sensitivity. Remember: Read access to secrets is still a breach. 2 / 7 2. What guidance does the article give for Anthropic Claude MCP security? 1. Use only public MCP endpoints for convenience 2. Allow all MCP tools without restriction 3. Host MCP servers internally – authenticate via mTLS or API keys – monitor unusual tool sequences 4. Disable MCP entirely as it cannot be secured Correct! Why: The article recommends hosting MCP servers internally and authenticating calls via mTLS or API keys and monitoring for unusual tool sequences. Context: Local MCP servers have file system access to the host machine making careful implementation review essential. Remember: Internal hosting – strong auth – watch for anomalies. 3 / 7 3. What fail-safe mechanisms does the article recommend for plugin containment? 1. Rate limits – maximum actions per session – time-bound execution – automatic disablement on anomaly 2. Only manual review of each plugin action 3. Complete isolation with no network access 4. Weekly security audits by external consultants Correct! Why: The article recommends rate limits on invocations plus maximum allowed actions per session plus time-bound execution windows plus automatic plugin disablement on anomaly detection. Context: These mechanisms prevent catastrophic damage even when plugins are compromised. Remember: Limit – Cap – Time-bound – Auto-disable. 4 / 7 4. According to the article – what approach is better for plugin security? 1. Equal numbers of plugins regardless of permissions 2. Only plugins that require administrator approval 3. Many plugins with broad access for maximum functionality 4. Fewer plugins with tight permissions rather than many plugins with broad access Correct! Why: The article states that fewer plugins with tight permissions is better security than many plugins with broad access. Context: Each unnecessary plugin or permission is potential exposure – enabling all plugins for convenience means maximum attack surface. Remember: Less is more for plugin security. 5 / 7 5. What does OWASP recommend regarding how plugins should be treated? 1. Treat plugins as untrusted with strict input validation and least privilege 2. Trust all plugins from official marketplaces 3. Only use plugins developed in-house 4. Disable all plugins by default with no exceptions Correct! Why: OWASP recommends treating plugins as untrusted while implementing strict input validation and applying least privilege. Context: This defensive posture acknowledges that plugins may be malicious or vulnerable and limits potential damage. Remember: Untrusted until proven otherwise. 6 / 7 6. What are the three main plugin threat vectors identified in the article? 1. Authentication – authorization – accounting 2. Malicious plugins – compromised plugins – exploited plugins via prompt injection 3. Encryption – hashing – tokenization 4. Network attacks – application attacks – physical attacks Correct! Why: The article identifies plugins that can be the attacker (malicious) or the victim (compromised) or the weapon (exploited via prompt injection). Context: Defense must address all three vectors because each requires different security controls. Remember: Malicious – Compromised – Exploited. 7 / 7 7. What makes AI plugin security fundamentally different from traditional integrations? 1. Traditional integrations are always less secure than AI plugins 2. The AI makes decisions about which plugins to call rather than humans explicitly choosing actions 3. Traditional integrations cannot connect to external systems 4. AI plugins require more bandwidth than traditional integrations Correct! Why: With AI plugins the AI chooses which actions to take rather than humans explicitly choosing – and the AI can be manipulated through prompt injection. Context: This decision flow creates unique risk because attackers can influence the AI interpretation to trigger unintended plugin actions. Remember: The AI decides – and the AI can be manipulated. Your score isThe average score is 0% Restart quiz Download PDF Please leave this field empty๐ The AI Security Manager's Newsletter Weekly insights on AI risk management, EU AI Act compliance, and practical security strategies. We donโt spam! Read our privacy policy for more info. Thank you! Please check your inbox to confirm your subscription.
