Free tools Windows power users keep installed
One-click scans. No signup required.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Stop workplace AI leaks with a layered program: approve specific tools and uses, classify and minimize data, enforce identity-aware retrieval, scan every prompt and context boundary, restrict connectors, monitor activity, and rehearse response. A promise that a provider does not train on your prompts cannot correct overshared files, unsafe memory, or a copied secret.
Why workplace AI leaks happen
A prompt is a data transfer. Microsoft advises treating a prompt as equivalent to sending information to an external service when the service has not been approved. An employee can paste a password, private key, source-code fragment, customer record, health detail, financial record, legal document, or security log while asking for a summary or troubleshooting help. The request may look harmless, but the content has already crossed a trust boundary.
Exposure also occurs without a user deliberately pasting a sensitive value. An assistant can assemble context from conversation history, persistent memory, retrieved document snippets, vector stores, tool output, caches, logs, or agent state. If a retention rule, tenant boundary, or authorization check is wrong, a later answer can disclose information to someone who should not receive it.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Authorization remains the decisive control for connected assistants. A copilot that inherits a user’s identity, permissions, sensitivity labels, retention rules, and audit policies is not a substitute for correcting the underlying repository. If a shared drive, SharePoint site, ticket queue, code repository, CRM, or knowledge base is overshared, an assistant may make that overshared content easier to find. Retrieval must enforce the requesting identity’s source-system permissions at query time.
#1 Best Overall
- POWERFUL SECURITY KEY: The YubiKey 5C NFC is the most versatile physical passkey, protecting your digital life from phishing attacks. It ensures only you can access your accounts
- WORKS WITH 1000+ ACCOUNTS: Compatible with popular accounts like Google, Microsoft, and Apple. A single YubiKey 5C NFC secures 100+ of your favorite accounts, including email, password managers, and more
- FAST & CONVENIENT LOGIN: Plug in your YubiKey 5C NFC via USB and tap it, or tap it against your phone (NFC), to authenticate. No batteries, no internet connection, and no extra fees required
- MOST SECURE PASSKEY: Supports FIDO2/WebAuthn, FIDO U2F, Yubico OTP, OATH-TOTP/HOTP, Smart card (PIV), and OpenPGP. That means it’s versatile, working almost anywhere you need it
- PRIMARY & SPARE KEYS: Just like having a spare house key, we recommend buying two YubiKeys - one for daily use and one as a spare. That way you’ll never get locked out of your accounts
Enterprise commitments reduce particular risks but do not make every workflow safe. Microsoft documentation describes controls such as encryption, tenant isolation, inherited policies, auditing, and a commitment not to train foundation models on prompts, responses, or Microsoft Graph data under the documented enterprise terms. Those protections still depend on the selected product, plan, feature, connector, retention setting, and tenant configuration.
Set a policy employees can actually follow
An acceptable-use rule should answer three questions before anyone opens an assistant: which tools are approved, which tasks are approved, and what information may be entered. Write the answers in plain language and publish them where work happens, not only in a security handbook.
Define approved tools and tasks
Maintain a current list of sanctioned chat interfaces, coding assistants, APIs, plug-ins, browser extensions, and agent platforms. Identify the owner, business purpose, data boundary, enabled connectors, retention period, and escalation contact for each one. Treat an embedded assistant inside another application as a separate system until its data flow and controls have been verified.
Specify permitted tasks such as drafting from public material, transforming synthetic examples, or summarizing documents that the requester is already authorized to access. Prohibit unreviewed use for decisions that affect employment, customers, safety, finances, or legal rights unless the responsible function has approved the workflow and a human remains accountable.
State prohibited inputs and safe substitutes
Tell employees not to enter credentials, private keys, authentication tokens, customer records, government identifiers, health or financial information, unreleased legal material, confidential source code, or security telemetry unless a documented exception covers the exact tool and task. Ask them to remove secrets, replace personal values with synthetic data, and redact identifying fields before requesting assistance.
Explain that an uploaded file, connector result, generated answer, and outbound action are data-handling events just like a typed prompt. Employees should verify an answer before sharing it because generated text can contain confidential context, fabricated claims, or an inappropriate recipient.
Create an exception path
Route exceptions through security, privacy, and legal review. The request should identify the data classes, tool and plan, connector scopes, retention and deletion behavior, people who can access the result, and the business justification. Record an expiry date so a temporary exception does not become permanent access.
Provide a sanctioned alternative that is available and useful. If the approved tool is slow, unavailable, or unable to handle a common task, staff will predictably move the work to an unsanctioned service. Training should include short examples of safe redaction and a clear method for reporting an accidental disclosure without fear of punishment for promptly reporting it.
Rank #2
- 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.
Inventory systems and classify information
Start with an inventory rather than a product promise. Include sanctioned and unsanctioned chat tools, API calls in applications, coding and analytics assistants, retrieval-augmented generation systems, autonomous agents, browser extensions, and assistants embedded in business software.
Map each system’s data path
For every system, document what it can read, write, remember, retrieve, export, or call. Trace the path from browser or API input through preprocessing, model context, retrieval, memory, logging, output filtering, and outbound actions. Record whether data is sent to a provider, stored in a tenant, copied to a vector index, written to a cache, or passed to another connector.
Include non-obvious state. Conversation history, long-term memory, vector embeddings, tool responses, agent plans, evaluation traces, debug logs, and support exports can all contain the original sensitive value or enough detail to reconstruct it.
Use practical data classes
Classify information as public, internal, confidential, restricted, or regulated, then map the classes to handling rules. Mark credentials, private keys, customer records, health and financial data, source code, legal material, and security telemetry as prohibited or tightly controlled inputs unless an approved exception exists. Classification should travel with a document and with derived artifacts, including snippets, embeddings, summaries, and generated files.
Assign an accountable owner to each data store and connector. An inventory that lists only models and vendors misses the repositories that actually contain the information an assistant can reveal.
Repair permissions before enabling copilots or retrieval
Review SharePoint sites, file drives, ticketing systems, code repositories, CRM records, and knowledge bases before indexing them. Remove broad links, stale group membership, inherited access that no longer matches job roles, and files whose sensitivity exceeds the purpose of the index.
Test authorization at retrieval time
A retrieval-augmented system should pass the requesting identity to the search layer and enforce document-level authorization for every query. Filtering only when a document is first indexed is insufficient: permissions change, users leave teams, and a previously private file may become restricted.
Recommended Free Tools
Test with two ordinary accounts and an administrator account. Confirm that a user cannot retrieve a document, snippet, embedding-derived passage, citation, or summary that the source system would deny. Test direct questions, paraphrased questions, bulk searches, and requests that ask the assistant to reveal its hidden context.
Rank #3
- POWERFUL SECURITY KEY: The YubiKey 5 NFC is the most versatile physical passkey, protecting your digital life from phishing attacks. It ensures only you can access your accounts
- WORKS WITH 1000+ ACCOUNTS: Compatible with popular accounts like Google, Microsoft, and Apple. A single YubiKey 5 NFC secures 100+ of your favorite accounts, including email, password managers, and more
- FAST & CONVENIENT LOGIN: Plug in your YubiKey 5 NFC via USB and tap it, or tap it against your phone (NFC), to authenticate. No batteries, no internet connection, and no extra fees required
- MOST SECURE PASSKEY: Supports FIDO2/WebAuthn, FIDO U2F, Yubico OTP, OATH-TOTP/HOTP, Smart card (PIV), and OpenPGP. That means it’s versatile, working almost anywhere you need it
- PRIMARY & SPARE KEYS: Just like having a spare house key, we recommend buying two YubiKeys - one for daily use and one as a spare. That way you’ll never get locked out of your accounts
Control indexes and derived copies
Separate indexes by tenant and, where needed, by business unit or sensitivity class. Apply deletion and permission changes to embeddings, caches, summaries, and backup copies. Do not assume that removing a source document immediately removes every derived representation.
For assistants that inherit identity and sensitivity labels, verify the actual configuration and license or plan. A product’s ability to honor a policy does not mean the policy is enabled for every connector or feature.
Isolate identity, memory, and tools
Design separate context boundaries for each user, tenant, session, task, agent, and retrieval scope. The narrower the boundary, the less a prompt-injection mistake or authorization defect can expose.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsKeep secrets out of model-visible context
Store tool credentials, system prompts, private keys, and sensitive operational state in a controlled secret-management service rather than in prompts, memory, or ordinary files. Give the model an opaque capability or narrowly scoped operation instead of the credential itself. Use managed identities or scoped service principals, short-lived tokens, and explicit read and write permissions.
Partition memory and caches
Give each user and tenant separate memory namespaces. Define what may be remembered, how long it may remain, who may delete it, and whether a user can inspect and correct it. Disable persistent memory for high-risk workflows unless there is a clear business need and a tested deletion process.
Apply the same isolation to response caches, conversation exports, evaluation datasets, prompt-trace systems, and agent state. A cache key that omits tenant, user, or authorization context can return one person’s answer to another person.
Constrain agent actions
Default agents to read-only access. Separate read and write tools, allowlist destinations, restrict network egress, and require human approval for external messages, record changes, purchases, permission changes, or deletion. Make the approval screen show the exact data and action, not merely a generic “run” button.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Put DLP checks at every boundary
Data-loss prevention must cover the entire path, not just the chat box. A secure proxy or gateway can provide a consistent inspection point for browser and API traffic, but controls inside the assistant and at the destination remain necessary.
Rank #4
- 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.
Inspect incoming content
- Scan typed prompts, pasted text, file uploads, images processed with optical character recognition, and API payloads.
- Detect credentials, private keys, government identifiers, customer records, source code, health information, financial information, and other regulated data.
- Block or quarantine high-confidence matches; redact or tokenize fields when the task can proceed safely; route exceptional cases for approval.
Inspect retrieved and remembered context
Scan snippets returned by search, records written to memory, vector-store ingestion, tool output, and agent state. Retrieval can reintroduce a secret even when the user typed a safe prompt. Memory-write scanning is especially important because a one-time mistake can become a durable disclosure.
Inspect outputs and actions
Check generated responses, downloaded files, citations, and outbound tool calls for sensitive markers and unauthorized destinations. A response containing a customer number may require redaction even when the original prompt passed inspection. For high-risk actions, hold the request until an authorized person reviews the proposed content and target.
Make policy decisions explainable
Log whether content was allowed, redacted, blocked, quarantined, or approved, along with the policy rule and accountable reviewer. Avoid storing the unredacted secret in a debug log merely because a DLP rule detected it.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Evaluate providers and connectors before approval
Request current contractual and technical answers for each service and configuration. Do not treat a general enterprise statement as proof that every model, plan, plug-in, region, or agent has the same behavior.
| Axis | Questions to ask |
|---|---|
| Data use | Are prompts, responses, files, and connector data used for model training? Can the setting change by plan or feature? |
| Isolation | Are tenants, users, agents, sessions, memory, and vector indexes isolated? |
| Authorization | Does retrieval enforce source-system permissions at query time? |
| DLP | Can the service inspect and block prompts, uploads, outputs, memory writes, and tool calls? |
| Retention | What is retained, for how long, where, and how is deletion verified? |
| Audit | Are interactions, administrator actions, retrievals, and policy decisions exportable to a SIEM? |
| Integrations | Which connectors, plug-ins, browsers, and outbound write actions are allowed? |
| Safety operations | Are prompt-injection defenses, human approval, testing, and incident response supported? |
Review legal and operational terms
Examine the provider’s data-processing agreement, retention and deletion terms, training-use language, data-residency commitments, subprocessor list, breach-notification deadlines, and support-access model. Confirm how a tenant export is handled, how deletion is verified, and what happens when a feature is retired or a plan changes.
Limit connectors by default
Allow only the connectors required for a documented use case. Separate read-only and write-capable integrations, restrict destinations, and review permissions after every product or workflow change. A connector that can send email, edit a record, open a ticket, or call a deployment system deserves the same scrutiny as a privileged application.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Strengthen identity and endpoint controls
Strong identity controls prevent an attacker from using an employee’s approved assistant session to extract data. Require single sign-on, phishing-resistant multifactor authentication where supported, device health checks for sensitive workflows, and rapid revocation when a device or account is compromised.
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 →A FIDO2 security key is a concrete phishing-resistant option for account access. It complements DLP: the key helps prevent account takeover, but it cannot stop an authenticated employee from pasting a secret into an unsafe prompt. Use least-privilege roles for administrators, separate administrative accounts from daily accounts, and review privileged access regularly.
Best Value
- FIDO2 & Passkey Ready: Business-ready and FIDO2 L1 certified. This key is supported by major management suites and is ideal for both individual and enterprise deployment. Works seamlessly with Gmail, Facebook, GitHub, Dropbox, Coinbase, and more.
- Universal Connectivity (USB-A ): Features a built-in USB-A connector—simply unfold the key and plug it into your compatible PC or laptop for seamless authentication on the go.
- Dedicated Manager App: Use the Thetis Manager App for the initial hardware PIN setup. Setting the PIN on the device first ensures a smooth registration process. Once the PIN is configured, you can begin registering the key across your favorite FIDO2-compatible online services.
- Ultra-Durable & Portable: Featuring a rotating metal cover, this key is water, crush, and tamper-resistant. It fits easily on a keychain and requires no batteries or network connectivity.
- Check FIDO2 compatibility before purchase - Known limitations: ID Austria is not supported (requires FIDO2 Level 2). Windows Hello login only works with Windows Enterprise editions that support Entra ID, and NFC is NOT supported.
On managed endpoints, control browser extensions, clipboard handling, downloads, local prompt history, and screen-capture exposure according to the sensitivity of the work. Endpoint controls should support the approved workflow rather than forcing staff to use personal devices or unsanctioned services.
Log, detect, and respond
Send prompt and response metadata, retrieval events, memory writes, tool calls, policy decisions, and privileged-access events to a monitored logging system. Retain enough context to investigate while applying the same minimization and access controls used for other sensitive logs.
Alert on behavior that indicates abuse or misconfiguration
- Cross-user or cross-tenant context access.
- Repeated attempts to extract secrets, hidden prompts, system instructions, or restricted records.
- Unusual bulk retrieval, rapid enumeration of repositories, or large downloads.
- Outputs containing credential patterns, private keys, government identifiers, or other sensitive markers.
- New connectors, changed retention settings, disabled DLP rules, or unexpected administrator access.
Test before production and after changes
Run adversarial tests for direct and indirect prompt injection, retrieval authorization, memory leakage, connector misuse, cache isolation, and output filtering. Repeat tests after a model, agent, connector, data source, plan, or policy changes. Include ordinary user accounts, privileged accounts, revoked accounts, and concurrent sessions.
Use a breach playbook
Prepare a sequence for credential rotation, token invalidation, session revocation, connector disablement, affected-data scoping, evidence preservation, notification, and recovery. Decide in advance who can stop an agent, who approves external notification, and how to distinguish a blocked attempt from a confirmed disclosure. After containment, remove exposed values from memory, caches, indexes, logs, and downstream systems where feasible, then document the control change that prevents recurrence.
Common failure modes and practical fixes
“The provider does not train on our data, so we are safe.”
Training use is only one exposure. A wrong connector permission, persistent memory entry, support export, or recipient address can disclose data without any model-training event. Verify isolation, retention, access, and outbound actions separately.
“We blocked public chat sites, so the problem is solved.”
Employees may still use an approved assistant incorrectly, paste secrets into an IDE plug-in, or encounter an embedded agent in a business application. Cover browser, API, desktop, mobile, plug-in, and embedded paths, and provide a usable sanctioned alternative.
“The source repository is private, so retrieval is safe.”
Private to whom is the relevant question. Stale groups, inherited permissions, broad links, and service accounts can make confidential content retrievable. Test authorization with multiple identities and inspect derived indexes and caches.
“We scan prompts but not memory or outputs.”
A safe prompt can retrieve a secret, and an allowed response can be copied into a ticket, email, or file. Inspect retrieval, memory writes, generated output, and outbound actions as well as input.
“An administrator can see everything for troubleshooting.”
Unrestricted support access creates another disclosure path. Use audited, time-limited elevation, minimize displayed content, and require a business reason for privileged inspection.
A defensible approval checklist
Approve a workplace AI tool only when the owner can answer these questions with current evidence:
- Which users, tenants, agents, sessions, memory stores, and indexes are isolated?
- Which data classes may enter prompts, files, retrieval, memory, logs, and outputs?
- Does every retrieval enforce source permissions at query time?
- Can DLP block or redact prompts, uploads, retrieved context, memory writes, outputs, and tool calls?
- What is retained, where, for how long, and how is deletion verified?
- Are encryption, MFA, network egress, connector allowlists, and privileged-access auditing enabled?
- Can interaction and policy logs reach the organization’s monitoring system without exposing the secret they are meant to protect?
- Have prompt injection, indirect injection, cross-user retrieval, memory leakage, cache isolation, and connector misuse been tested?
- Is there a named owner, an expiry or review date, and an incident playbook?
NIST’s National Cybersecurity Center of Excellence states that an organization must protect its information from unauthorized access and disclosure. Applied to workplace AI, that obligation follows the data through prompts, context, memory, retrieval, output, and action. Recheck permissions, connector scopes, retention, and provider terms whenever a product, plan, model, agent, or data source changes.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
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.
