Before an AI agent can act in organizational systems, define what it may do and access, restrict its permissions and execution environment, and set clear human-approval gates. Test the agent with its actual tools and identities, monitor its activity, retain records needed to investigate consequential actions, and prepare a way to pause it and revoke access. The right controls depend on the agent’s capabilities and the impact of its mistakes.
Start by defining the agent’s operating boundary
Write down the agent’s approved purpose and tasks before connecting it to live systems. Specify what it must not do, which data it may access, which tools it may call, and where it is allowed to operate. Name the owner responsible for the agent and identify who can change its instructions, tools, or permissions.
As an Amazon Associate I earn from qualifying purchases.
This boundary is the basis for deciding whether a later request or action is authorized. If the approved tasks, data, or systems are vague, it will be difficult to set useful permissions or determine whether the agent has crossed a line.
Recommended Free Tools
Apply these controls before granting live access
-
Limit accounts, credentials, and data access
Give the agent only the accounts, data, and tool scopes needed for its approved tasks. Avoid sharing a broad human or administrator account with the agent. Where practical, separate credentials by environment or task, protect secrets, and make access straightforward to revoke. Define the data the agent can read separately from the data it can change.
#1 Best Overall
Yubico - Security Key C NFC - Basic Compatibility - Multi-Factor authentication (MFA) Security Key and passkey, Connect via USB-C or NFC, FIDO Certified- 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.
NIST’s agent-security materials identify constraining and monitoring the extent of an agent’s access as a deployment concern. The specific account design and permission scopes are implementation choices for the organization, not a universal NIST-prescribed configuration.
-
Constrain code execution and external actions
Restrict code execution to approved environments. Use a sandbox, monitoring, or an approval step where unrestricted execution would create unacceptable risk. Limit which tools and destinations the agent can reach; allowlists and usage limits can help keep actions within the approved boundary. Consider separately whether an agent may send messages, modify records, deploy code, or initiate transactions.
Choose restrictions according to the task: a read-only assistant and an agent that can change production systems do not need the same action surface. NIST’s Control Overlays for Securing AI Systems (COSAiS) materials describe ways to select, adapt, and supplement controls for different technologies and operating environments; they are guidance to tailor, not a single configuration to switch on.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsSpecial offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy. -
Set human-approval gates for consequential actions
Decide which actions the agent may complete independently and which must wait for a person. Consider requiring review when an action has significant impact, unclear authorization, external consequences, financial effects, changes access, or is difficult to reverse. For lower-impact tasks, bounded permissions and monitoring may be proportionate.
Rank #2
Yubico - Security Key NFC - Basic Compatibility - Multi-Factor Authentication (MFA) Key, Connect via USB-A or NFC, FIDO Certified- POWERFUL SECURITY KEY: The Security Key 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 NFC secures 100 of your favorite accounts, including email, password managers, and more.
- FAST & CONVENIENT LOGIN: Plug in your Security Key NFC via USB-A 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.
There is no universal approval threshold established for every agent. Set one based on the action’s consequences, reversibility, and the organization’s obligations, and make the gate operational: identify who approves, what information they see, and how the agent behaves while approval is pending.
-
Test the complete deployment, not just the model
Evaluate the combination that will actually run: model, instructions, tools, identities, data, permissions, and workflow. Check both that approved tasks work and that the agent respects access restrictions and approval gates. Include cases where a request is unauthorized or ambiguous, as well as cases where a tool or service fails.
Repeat testing when the model version, tools, permissions, data, or workflow materially change. NIST’s AI Risk Management Framework (AI RMF) treats testing and evaluation as lifecycle work that extends into deployment and use, rather than a one-time pre-launch check.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy. -
Monitor activity and retain useful records
Monitor tool use, access, errors, and attempted boundary crossings. Keep enough records to reconstruct consequential actions and support incident review, while following applicable privacy and data-retention rules. Decide in advance who reviews alerts and what conditions require escalation; telemetry without an owner or response path may not help contain a problem.
Rank #3
The appropriate events to log and how long to retain them depend on the deployment. NIST’s January 12, 2026 notice seeking input on agent security identified constraining and monitoring agent access among the possible deployment interventions; it did not establish a universal monitoring or retention specification.
-
Prepare to pause, contain, and recover
Identify who can pause or disable the agent, revoke its credentials, and contain its execution environment. Define how the organization will handle an incident, including how it will assess actions already taken and restore affected systems or data where possible. Exercise the response path before granting broad access so people know how to intervene under pressure.
-
Assign an owner for ongoing review
Have an accountable owner review controls when the agent’s model, tools, data, users, or operating environment change, and when monitoring or testing reveals unexpected behavior. Revisit the approved boundary and permissions as the agent’s role evolves rather than assuming the initial deployment remains safe indefinitely.
DriversOutdated Drivers Are Slowing You DownPerformanceWindows Errors? Fix Them Before They SpreadDriversCrashes, No Sound, or Screen Glitches?Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
Use a risk framework to organize the work
NIST’s AI RMF 1.0, released January 26, 2023, is voluntary. Its four functions can organize control decisions and follow-up without turning the framework into a universal legal checklist. NIST’s AI RMF Playbook provides suggested actions based on the framework.
Rank #4
- 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.
| AI RMF function | How it can help with an agent |
|---|---|
| Govern | Establish ownership, accountability, and the process for approving or changing the agent’s role. |
| Map | Describe the agent’s context, tasks, data, tools, users, and potential impacts. |
| Measure | Evaluate performance and risks, including whether the deployed configuration respects its limits. |
| Manage | Apply controls, monitor issues, respond to incidents, and revisit decisions as conditions change. |
NIST’s AI RMF resources address trustworthiness across pre-design, design and development, deployment, use, and test and evaluation. COSAiS adds agent-specific single-agent and multi-agent use cases drawing on SP 800-53 controls. NIST describes these use-case materials as implementation guidance in development, not a finalized mandatory agent standard.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Compare deployment options by their risk and safeguards
When choosing between ways to deploy an agent, compare the actual capabilities and protections rather than relying on a general claim that one setup is “secure.” These are practical comparison dimensions, not a vendor ranking or a scored NIST benchmark.
| What to compare | Question to ask |
|---|---|
| Actions and systems | Which systems can each option reach, and what can it do there? |
| Data | How sensitive is the information it can read or change, and how broad is that access? |
| Autonomy and tools | How independently can it act, and how many tools or destinations can it use? |
| Impact and reversibility | What is the likely impact of an error, and how readily can the action be undone? |
| Approval, monitoring, and recovery | How strong are the human gates and monitoring, and can the organization intervene and recover? |
| Test evidence | What evidence shows that this configuration was tested in its intended environment? |
Account for evolving agent-security guidance
NIST describes agent systems as capable of autonomous decisions and actions with limited human supervision. Its agent-system descriptions include understanding context, reasoning, planning, adapting, and executing tasks. This means the control surface includes not only model behavior but also the software tools, permissions, data flows, and environment through which the agent acts.
NIST’s Center for AI Standards and Innovation (CAISI) announced an RFI on securing agent systems on January 12, 2026, asking for input on deployment interventions, including ways to constrain and monitor access. The comment deadline in the notice was March 9, 2026, and has passed. NIST’s May 18, 2026 analysis of responses reports broad agreement among respondents that agent security risks are novel and that traditional cybersecurity practices remain relevant but need adaptation. These are evolving materials; consult current NIST guidance and applicable requirements when making deployment decisions.
The NIST AI RMF is voluntary, and the sources do not establish a single approval threshold, testing depth, retention period, or legal obligation for every deployment. Those decisions depend on the agent’s capabilities, the data and systems involved, likely impacts, sector, jurisdiction, contracts, and organizational risk tolerance.
Quick Recap
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.
