The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →If an AI agent has made cloud changes you did not approve, treat the event as both a cloud-identity incident and an agent-behavior incident. Preserve evidence, contain the credentials and runtime involved, trace the principal’s API activity across services and regions, check for persistence, then remove the unauthorized access path and recover from a known-good state.
1. Preserve the alert and the evidence
Before deleting resources, rotating credentials, or restoring data, record what triggered the response and preserve the records needed to understand what happened. Start with the agent or workload identity, associated service account or role, affected cloud account or project, approximate time window, and suspicious actions.
As an Amazon Associate I earn from qualifying purchases.
- Save relevant audit, application, agent-runtime, and network records, including their time ranges and the systems they came from.
- Record affected resource names, identifiers, configurations, and current state. Where practical, capture snapshots or other evidence before changing or removing a resource.
- Keep a timeline of response actions, including who made each change and when. This helps distinguish the agent’s activity from containment and recovery work.
Google Cloud recommends backing up logs for affected resources for forensic analysis in its AI threat-finding response guidance. AWS likewise advises backing up resources that must remain available for investigation in its account-compromise guidance. Preserve evidence in a location and format that will not be overwritten by routine retention or cleanup.
Free tools Windows power users keep installed
One-click scans. No signup required.
2. Contain the identity, credentials, and runtime
Identify which principal made the suspicious calls and which credentials or sessions it could use. Containment should stop further unauthorized actions without assuming that pausing the agent or disabling one account invalidates every credential it has already obtained.
#1 Best Overall
- A FIDO security key with PUF technology provides a unique, hardware-rooted trust anchor that resists tampering and cyber attacks, offering stronger security than conventional designs.
- FIDO2 Certified Protection – Enjoy phishing-resistant security with FIDO2 certification, ensuring top-tier account safety across Windows, macOS, Linux, iOS iOS, Android and more.
- Easy to use & Portable – Designed with a compact USB-C interface, Clife key fits easily on your keychain for secure access anywhere. Simply plug in and authenticate with ease.
- Universal Compatibility – Works seamlessly with hundreds of FIDO2/U2F compliant services, including popular cloud, email, and social platforms.
- Backup recommended – To ensure continuous access, register a backup Clife security key as a spare in case your primary key is lost.
- Revoke, disable, or restrict the implicated credentials and principal using the provider’s current procedure. Check for both long-lived keys and short-lived tokens or sessions.
- If feasible, stop or isolate the implicated agent runtime while you assess whether doing so will interrupt legitimate work.
- Limit permissions for related identities when the evidence suggests they are part of the same access path.
Google’s guidance for compromised service accounts specifically calls for addressing both persistent key files and short-lived access tokens; revoking only one form may leave another usable. Follow the provider’s procedure for the identity type involved rather than assuming that a user suspension, agent shutdown, or key change has invalidated every credential: Google Cloud: Respond to compromised Google Cloud credentials.
Containment can disrupt legitimate workloads. Before deleting an identity or resource, determine whether services depend on it and, where time permits, prefer a reversible restriction while you establish scope.
3. Reconstruct what the agent did
Build the timeline from provider audit events tied to the relevant principal and session. Look beyond the first alert: trace actions before and after it, and expand the search across services and regions. An agent’s cloud activity is usually established through the identities and APIs it used; a conversation transcript alone does not show the full set of cloud-side effects.
Rank #2
- Protect accounts with USB-A & NFC 2FA security key. Hardware-based authentication blocks phishing, credential theft & unauthorized access across cloud, enterprise & personal platforms.
- FIDO2 Level 2 certified Security Key. TAA compliant and supports Apple ID, Microsoft Azure/Entra ID, AWS, Google, Facebook, Salesforce, DUO & more. Works with Chrome, Safari & Edge across major OS.
- Plug & play USB-A Security Key with NFC tap login. No software, drivers or batteries required. Works with Windows PC, MacBook, iPhone, Android & Chromebook for fast, secure authentication.
- Built with FIPS 140-2 Level 3 secure element for advanced encryption. Trusted by IT teams, healthcare, education & government for secure authentication and identity protection.
- IP68 waterproof, dustproof & crush-resistant design. Supports FIDO2, U2F, OTP, PIV, Mini Driver & smart card login. Durable USB security key for long-term enterprise and daily use.
- Search for resource reads as well as changes, creation, and deletion. Note the affected service, resource, action, timestamp, principal, and session where available.
- Correlate audit-event times with network-flow and application logs. Use agent or model-invocation records when they were enabled and are available.
- Check whether identity or access policy changed during the window. Those changes may explain how the agent gained access or how access persisted.
- Keep confirmed events separate from hypotheses. External content the agent processed, such as documents or logs, may be worth investigating as a possible prompt-injection route, but its presence does not prove that it caused the activity.
AWS recommends searching CloudTrail across regions and services for activity associated with the implicated role session, then correlating event times with VPC Flow Logs and application records in its CloudTrail investigation guidance. If prompt-and-response logging was not enabled, that source cannot supply content that was never recorded; missing conversation data is not evidence that no cloud activity occurred.
4. Establish the scope and look for persistence
Use the principal and timeline as starting points, not as the boundary of the investigation. Determine which data and resources the identity could reach, what it actually accessed, and whether another identity or resource was created or altered to preserve access.
- Look for unfamiliar agent instances, sessions, service accounts, agent identities, roles, policies, and access keys.
- Review compute resources, applications, storage, snapshots, and other changes associated with the principal or time window.
- Check whether logs are complete and trustworthy: consider whether logging was enabled for the relevant service and region, whether retention covers the period, and whether permissions or log settings changed.
- Record what remains uncertain because telemetry was unavailable, disabled, or outside retention.
Google Cloud’s AI-agent finding guidance points investigators to audit records for service-account creation, IAM policy changes, and calls associated with the principal under review: Discovery: AI Agent Service Account Self-Investigation. Its broader AI threat guidance also recommends checking for unfamiliar Agent Runtime instances, sessions, service accounts, and agent identities: Respond to AI threat findings.
Rank #3
- Protect accounts with USB-C & NFC 2FA security key. Hardware-based authentication blocks phishing, credential theft & unauthorized access across cloud, enterprise & personal platforms.
- FIDO2 Level 2 certified Security Key. Works with Apple ID, Microsoft Azure/Entra ID, AWS, Google, Facebook, Salesforce, DUO & more. Compatible with Chrome, Safari & Edge on all major OS.
- Plug & play USB-C Security Key with NFC tap login. No software, drivers or batteries required. Works with Windows PC, MacBook, iPhone, Android & Chromebook for fast, secure authentication.
- Built with FIPS 140-2 Level 3 secure element for advanced encryption. Trusted by IT teams, healthcare, education & government for secure authentication & identity protection.
- IP68 waterproof, dustproof & crush-resistant design. Supports FIDO2, U2F, OTP, PIV, Mini Driver & smart card login. Durable USB security key for long-term enterprise & daily use.
5. Remove the unauthorized path and recover
Once you have enough evidence to understand the access path and its operational impact, remove unauthorized permissions, credentials, and resources. Fix the agent, application, or integration path that allowed the actions; deleting the visible resource alone will not prevent a repeat if the underlying access remains.
Recommended Free Tools
- Remove or correct unauthorized policy grants, keys, identities, and resources identified during scoping.
- Constrain the agent’s permissions to the cloud actions and resources its task actually requires.
- Restore affected services or data from a known-good source, rather than from a state that may contain unauthorized changes.
- Validate that the restored service behaves as expected and that its required legitimate identities still work.
- Continue monitoring the relevant principals, services, and resources for renewed use or new changes.
Provider procedures differ, and removing credentials or resources can affect dependent workloads. Google’s compromised-credential guidance includes restoring service and reviewing resource access after credential response; AWS’s account-compromise guidance advises checking resources across regions. Use those provider-specific procedures for the affected environment rather than applying a single cross-cloud command sequence.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Where to look in each cloud environment
AWS
For generative-AI workloads, AWS names CloudTrail, CloudWatch, VPC Flow Logs, S3 data events, and Amazon Bedrock model-invocation logs (when the application uses Bedrock) as relevant investigation sources. The goal is to establish which resources were accessed, modified, created, or deleted: AWS methodology for incident response on generative AI workloads.
Rank #4
- FIDO2 + FIDO U2F certified and supported USB security key
- Supports Computers, Laptops, Tablets, and Mobile Devices with a USB-C port
- Works without downloading any drivers. Supported OS: Android, Chrome OS, Windows, MacOS, Linux
- Durable design made to last for a long time with everyday use. Water-resistant (IP67)
- Helps protect your accounts from phishing and other cyber-attacks. Prevents your devices from unauthorized use.
For account-compromise checks, AWS also recommends examining CloudTrail Event history for unsanctioned creation of access keys, policies, roles, or temporary credentials, checking resources across regions, and preserving resources needed for investigation: Resolve issues with unauthorized activity in AWS accounts.
Google Cloud
Use Cloud Logging and Security Command Center findings to investigate the relevant principal and resources. Confirm that the responders have the roles needed to view the audit and Data Access logs they need; incomplete access to logs can leave important activity out of view. Google’s response guidance covers service-account keys, unexpected resources such as VMs, applications, service accounts, and storage buckets, and post-containment access review: Respond to compromised Google Cloud credentials.
Google’s instruction for AI threat findings is: “Before you take any action, you should investigate the findings; assess the information that you gather; and decide how to respond.” See Respond to AI threat findings.
Best Value
- SECURITY KEY FOR ENTERPRISE ACCESS: Supports FIDO2 passkeys and U2F for secure authentication across enterprise IT systems.
- PHISHING-RESISTANT AUTHENTICATION: Enables passwordless login with secure on-device credential storage and PIN-based user verification.
- COMPATIBLE WITH ENTERPRISE SYSTEMS: Works with FIDO2, WebAuthn, and U2F across enterprise, cloud, and modern IT environments.
- DRIVERLESS FIDO2 AUTHENTICATION: FIDO2 works natively with modern browsers and platforms. No drivers required.
- USB AND NFC CONNECTIVITY: Supports authentication via USB-C and NFC. No batteries required.
Microsoft Entra agent identities
When the investigation involves an Entra agent identity, review risk-detection details, sign-in logs, and audit logs. Agent identity creation can appear in audit activity such as “Create user” or “Create service principal.” These records help investigate the identity; if Azure resources were changed, also consult the applicable resource logs. See Microsoft’s Manage agent identities in your organization.
Document what is known—and what is not
Close the response record with a timeline of confirmed actions, affected identities and resources, containment and recovery changes, and the validation performed. Mark hypotheses as hypotheses. State plainly which logs were unavailable, not enabled, or no longer retained, and which questions therefore remain unresolved. A gap in telemetry limits what can be concluded; it does not establish that the agent took no other action.
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.
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 errors
