AWS can help enforce defined rules around an AI agent and preserve useful operational evidence, but no AWS service certifies an application or organization as compliant with the EU AI Act. To govern an agent, combine configured policy and access controls with monitoring and deliberately designed logs. To assess the Act, classify the system by its intended purpose and context, identify your organization’s role, and apply the obligations and dates relevant to that case.
What AWS controls can—and cannot—do
AWS services address different points in an agent’s lifecycle. Bedrock Guardrails can evaluate inputs and model responses for configured safeguards. Amazon Bedrock AgentCore policy can apply configured guardrail checks to requests and suppress outputs when a configured condition is met. IAM and service permissions determine what an agent’s identity is actually allowed to do. GuardDuty AI Protection looks for specified suspicious patterns in relevant CloudTrail events. CloudTrail and service logs can help preserve an activity record.
As an Amazon Associate I earn from qualifying purchases.
These controls are complementary, not interchangeable. A content filter is not an authorization system; a detector is not a blocker; and a log is not proof that a legal duty has been met. AWS describes the operating model succinctly: “Security is a shared responsibility between AWS and you.” (AWS, Security, Guardrails, and Observability in Amazon Bedrock.) In practice, AWS supplies configurable services and infrastructure controls, while you must decide how they apply to your application, data, users, and legal obligations.
How to block an AI agent from calling a tool
Do not rely on a prompt telling the model not to use a tool. Enforce the boundary in the permissions and policy that govern the action, then add content safeguards where appropriate.
#1 Best Overall
- POWERFUL SECURITY KEY: The Security Key C NFC is the essential physical passkey for protecting your digital life from phishing attacks. It ensures only you can access your accounts.
- WORKS WITH 1000+ ACCOUNTS: Compatible with Google, Microsoft, and Apple. A single Security Key C NFC secures 100 of your favorite accounts, including email, password managers, and more.
- FAST & CONVENIENT LOGIN: Plug in your Security Key C NFC via USB-C and tap it, or tap it against your phone (NFC) to authenticate. No batteries, no internet connection, and no extra fees required.
- TRUSTED PASSKEY TECHNOLOGY: Uses the latest passkey standards (FIDO2/WebAuthn & FIDO U2F) but does not support One-Time Passwords. For complex needs, check out the YubiKey 5 Series.
- BUILT TO LAST: Made from tough, waterproof, and crush-resistant materials. Manufactured in Sweden and programmed in the USA with the highest security standards.
- Constrain the agent’s identity. Give its service role only the permissions needed for its intended tasks. AWS’s prerequisites for Bedrock Agents include configuring an execution role and permissions for the resources the agent uses; use that as a starting point, then narrow permissions to your actual tool and data scope. See Prerequisites for creating Amazon Bedrock Agents.
- Define the action boundary. Limit which APIs, resources, and operations the role can reach. If an action is consequential, design an approval or circuit-breaker path rather than granting broad access and expecting a model instruction to contain it.
- Configure content safeguards. Bedrock Guardrails supports configured checks such as content filters, denied topics, sensitive-information filters, and prompt-attack detection. Guardrails can be applied to Agents and Knowledge Bases, but the safeguards must be deliberately configured and validated. See Use cases for Amazon Bedrock Guardrails and How Amazon Bedrock Guardrails works.
- Apply AgentCore policy where it fits. AgentCore policy guardrails support checks for prompt attacks, content filters, and sensitive information. A configured policy can check requests or suppress outputs when a guardrail condition is met. Follow the documented IAM permissions and verify service and Region availability for your deployment; do not assume that a policy feature is available in every Region or that an unconfigured policy blocks anything. See Guardrails in policies – Amazon Bedrock AgentCore.
- Test allowed and denied paths. Exercise representative benign, disallowed, sensitive, and adversarial inputs; confirm that the intended policy fires and that the agent cannot perform a prohibited operation through another permission path. AWS describes Guardrails safeguards as model-powered and periodically updated, and recommends continued testing and validation (How Amazon Bedrock Guardrails works).
The key distinction is that IAM and the permissions behind a tool determine whether the action is authorized, while a configured policy or guardrail can evaluate defined requests and outputs. Treat content filtering as an additional layer, not a substitute for least privilege.
Choose the control by its job
| Control | Primary job | What it does not establish | Implementation consideration |
|---|---|---|---|
| IAM and service permissions | Limit what the agent’s AWS identity can access or invoke. | That a prompt or model response is safe, or that all legal obligations are met. | Scope roles and credentials narrowly around actual tools and resources. |
| Bedrock Guardrails | Evaluate configured inputs and responses for content, sensitive-information, denied-topic, or prompt-attack conditions. | That an agent cannot invoke an API its identity is allowed to call. | Configure, test, and validate the safeguards for the application. |
| AgentCore policy guardrails | Apply configured guardrail checks to requests and suppress outputs when conditions are met. | That every tool call is authorized or that the feature is available in every Region. | Check IAM requirements and current Region support before deployment. |
| GuardDuty AI Protection | Analyze relevant CloudTrail events for specified suspicious behavior. | Authorization to make a tool call, or a legal compliance determination. | Enable the feature and review current charges, which are based on the volume of CloudTrail data events analyzed. |
| CloudTrail and service logs | Preserve records useful for identifying activity, identities, and AWS operations. | A complete record of every prompt or response unless the relevant logging is configured. | Set scope, retention, integrity controls, reviewer access, and sensitive-data handling deliberately. |
Monitor suspicious activity separately from blocking it
GuardDuty AI Protection analyzes CloudTrail events from Bedrock, AgentCore, and SageMaker AI for patterns that include anomalous model invocations, cost harvesting, and prompt-injection findings associated with Guardrails detections. The feature must be enabled. Its findings can support investigation and response, but they do not authorize tool calls or prevent every harmful action.
Rank #2
- OTP Token in card format that provides secure remote access with strong authentication
- Easy to use and easy to carry, same size as a credit card
- Zero footprint; No software on end-user PCs
- Compliant to OATH open standard (time based - 6 digits)
- Expected battery life is 3 years or approximately 15,000 clicks
AWS bases GuardDuty AI Protection usage charges on the volume of CloudTrail data events analyzed. Check the current AWS pricing information before estimating costs; the documentation describes the feature and its prerequisites at GuardDuty AI Protection.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Design evidence without creating a new data risk
Start with the questions an investigator, reviewer, or auditor will need to answer: who or what invoked the model, which identity was involved, which AWS operations occurred, and what response or escalation followed. Use CloudTrail and relevant service logs to preserve the activity records needed for those questions, and decide in advance who may review them and how long they are retained.
Rank #3
- Ultra-Compact FIDO2 Security Key - Plug-and-stay or carry on a keychain. This USB-A hardware security key offers portable, always-on protection for desktop and mobile use. (Item Size: 0.75 X 0.74 IN x 0.25 IN)
- USB-A Hardware Key for All Devices - Works with USB-A ports on PC, Mac, Android, and other laptop/notebook device. Enables secure, cross-platform login with FIDO2.0 passkey support.
- FIDO Certified Security Key - Meets FIDO and FIDO2 standards. Works with Google, Microsoft, GitHub, Dropbox, and more. Please check service compatibility before purchase.
- Passwordless Login with Passkey - Supports passkey login via WebAuthn and CTAP2. Enjoy password-free sign-ins where supported. Not all websites or services currently support passkeys.
- Advanced Multi-Factor Authentication - Offers 200 FIDO2 passkey slots and 50 OATH-TOTP slots. Strong, flexible 2FA/MFA support across various apps and authentication platforms.
Do not assume an activity log contains the full conversational record. If you enable invocation logging, Bedrock Guardrails documentation notes that blocked content can appear in plain text in those logs. Prompts and outputs may also carry personal data, credentials, or other sensitive information. Decide whether that content should be logged at all, restrict access, and apply retention and protection appropriate to the use case. See How Amazon Bedrock Guardrails works.
CloudTrail compliance documentation explains AWS’s compliance validation materials for the service (Compliance validation for AWS CloudTrail). AWS Artifact provides third-party audit reports for AWS services. Those reports can inform your assessment of AWS service controls; they do not replace evidence about your own agent’s design, risk assessment, oversight, monitoring, or incident handling.
Rank #4
- Feature: Material is four strong magnets in white plastic house
- Functions: It is used for displaying your stuffs so that it beautifies and saves your space while it prevents your retail items from missing.Key unlocks your hook lock as security magnetic key ,it meets many purposes.It is suitable for any specific security hook like 6"7"8"peg&slat wall hook& other usages.
- To use:You put it on the correct position when two tabs are in line ,then you slide it, so you unlock articles
- Warranty: Erase electronic data off most devices. SO BE CAREFUL PLACING OR STORING ELECTRONICS NEAR,To keep them away from your wallet avoid damaging your credit pinch fingers slamming together or grab up metallic objects
Classify the AI system and identify your role
The European Commission describes the AI Act as a risk-based framework spanning prohibited practices, high-risk systems, transparency requirements, and minimal- or no-risk applications. “Agent” describes an architecture, not an AI Act risk category. Classification depends on intended purpose and context, and obligations depend in part on whether an organization is acting as a provider, deployer, importer, distributor, or another regulated actor. The Commission’s overview is the starting point for that assessment: AI Act.
Free tools Windows power users keep installed
One-click scans. No signup required.
For systems classified as high-risk, the Commission identifies requirements including risk management, data governance, technical documentation, logging and traceability, information to deployers, human oversight, and accuracy, robustness, and cybersecurity. Which requirements apply, and to whom, depends on the classification and role. Do not infer a legal conclusion from the use of an AWS service or the word “agent”; use current Commission material and qualified counsel for the specific system.
Best Value
- Security Key : Protect your online accounts against unauthorized access by using FIDO2 and U2F authentication with T110. It's the world's most protective security key that works with windows, Mac OS, Linux as well as Chrome, Firefox, Edge and many other major browsers.
- Certified with the new FIDO2 standard, T110 provides the benefit of fast login and strong protection against phishing, account takeover as well as many other online attactks.
- Works with : Bank of America, Github, Google, Microsoft, DUO, Twitter, Facebook, Dropbox, Apple, ebay, BINANCE, mor and more.
- Fits USB-A port : Insert the T110 security key into the USB-A port of each service and log in conveniently with one touch
- For the driver download and user guide, please visit TrustKey Solutions Home support page.
General-purpose AI model obligations are a separate track from the classification of an application built with a model. The Commission lists technical documentation, a copyright policy, and a public summary of training content among obligations for GPAI model providers; systemic-risk models have additional duties. Whether a given organization has provider obligations depends on its role and the facts, not merely on its choice of cloud platform. See General-purpose AI obligations under the AI Act.
Use the current EU AI Act dates for the relevant obligation
The following dates reflect European Commission pages reviewed on 4 October 2026. They are application dates, not statistics, and different obligation tracks do not share one start date.
| Date | What the Commission says applies | Scope note |
|---|---|---|
| 2 February 2025 | Prohibited-practice rules and AI literacy provisions began applying. | The Commission overview describes an additional prohibition concerning non-consensual intimate material and child sexual abuse material from 2 December 2026. |
| 2 August 2025 | Governance rules and GPAI provider obligations began applying. | GPAI provider duties are a separate track; systemic-risk models have additional obligations. |
| 2 August 2026 | Enforcement powers for the AI Office and national authorities apply. | See the Commission’s enforcement framework. |
| 2 December 2027 | Annex III high-risk system rules apply. | Applicability depends on whether the system falls within the relevant classification. |
| 2 August 2028 | Rules apply to high-risk AI systems embedded in regulated products. | Applicability depends on the product and system falling within the relevant scope. |
Check the Commission’s current AI Act overview, enforcement page, and GPAI obligations page when planning a release or compliance review, especially where an obligation depends on system classification or organizational role.
Recommended Free Tools
Quick Recap
Build a defensible governance plan
- Document the use: record intended purpose, users, affected people, deployment context, and the agent’s tools and data access.
- Determine scope and responsibility: assess the applicable AI Act category and your organization’s role before selecting obligations and deadlines.
- Map each risk to an enforcement point: use narrowly scoped permissions for tool access, configured guardrails for content conditions, and human approval or a circuit breaker for consequential actions.
- Test what can fail: verify denied requests, allowed actions, alternate access paths, policy behavior, and operational recovery procedures.
- Plan detection and response: decide which CloudTrail events and GuardDuty findings matter, who reviews them, and what happens when an alert fires.
- Set evidence and data rules: decide what is logged, whether prompts or responses are included, who can access logs, and how retention and integrity are managed.
- Keep customer-side evidence: assemble records for system design, risk assessment, human oversight, monitoring, and incident handling alongside any AWS service audit reports.
- Reassess changes: revisit permissions, policies, tests, evidence, Region support, and legal classification when the agent’s purpose, model, tools, data, or deployment changes.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
