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1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsiTechGuides is reader-supported.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
Could an AI agent cause a data breach? Yes—but the claim that your next breach will be caused by one is not established. The credible risk is that an autonomous software agent can process hostile content and take actions through connected tools, sometimes with more access than its task requires. Government guidance describes several ways that can go wrong, while recent industry surveys report agent-related incidents and control gaps. Those surveys do not establish how many confirmed breaches agents have caused.
What does “AI employee” mean here?
“AI employee” is a metaphor, not a technical or legal category. The relevant systems are software agents: systems that use models, data, and software tools to pursue tasks with some degree of autonomy. Depending on their design, they may read files or email, query business systems, or take actions through connected applications.
That combination changes the security question. A model may generate an unsafe instruction, but the consequences depend on what the surrounding software lets the agent see and do. An agent with access to sensitive records and the ability to send messages presents a different risk from one that can only draft text for a person to review. NIST’s February 5, 2026 concept paper frames agent identity and authorization as important areas of work: New Concept Paper on Identity and Authority of Software Agents.
How can prompt injection hijack an AI agent?
In indirect prompt injection, an attacker hides malicious instructions in content an agent is expected to handle—such as an email, file, or website. The content can look like ordinary task material. If the agent’s design combines developer instructions with that incoming material in a way that lets the hostile text influence its behavior, the agent may follow the attacker’s instructions instead of its intended task.
#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.
A plausible attack chain
- The agent is given a useful task. For example, it is asked to summarize incoming messages or find information in a set of documents.
- It encounters attacker-controlled content. A message or page contains instructions intended for the agent, not just information relevant to the user’s request.
- The content redirects the agent. If the agent treats those instructions as authoritative, it may attempt to use its available tools in an unintended way.
- Its permissions determine the possible impact. An agent with access to sensitive data or consequential actions could expose information or disrupt operations; a narrowly scoped agent may have less ability to do harm.
This is a mechanism, not a claim that every agent is vulnerable or that this chain has caused a particular reader’s next breach. In a January 17, 2025 technical blog, NIST’s Center for AI Standards and Innovation (CAISI) described agent hijacking as a form of indirect prompt injection and explained how malicious instructions in data an agent ingests can redirect it.
What NIST’s evaluations show—and what they do not
CAISI tested agents powered by Claude 3.5 Sonnet, released in October 2024, in AgentDojo environments and custom scenarios. In three added scenarios—remote code execution, database exfiltration, and automated phishing—the team said it was “frequently able to induce the agent to follow the malicious instructions.” That is a result under specified evaluation conditions. It is not a real-world breach rate, a finding about all agents, or a measurement of today’s systems in every deployment.
CAISI recommends improving shared evaluation frameworks, adapting red-team tests as attacks change, assessing behavior on specific tasks, and measuring attack success across multiple attempts. A single successful or unsuccessful test is not enough to characterize risk across tasks and deployments.
Rank #2
- Hardware-Rooted Security with PUF Technology – PUFido Drive Clife Key uses Physical Unclonable Function technology to generate a unique, hardware-based identity that cannot be duplicated, delivering stronger resistance against tampering and cyber attacks than conventional security keys.
- FIDO2 Certified Phishing-Resistant Protection – Fully compliant with FIDO2/U2F standards, enabling secure passwordless login and two-factor authentication to help protect accounts from phishing and credential theft.
- Security Key + Flash Drive in One Device – Combines a FIDO security key with a built-in USB flash drive, allowing you to carry files and a hardware authentication key together in a single compact device.
- Easy to Use & Portable – Compact USB-C design fits easily on a keychain or in a pocket. Simply plug in the Drive Clife Key to authenticate or access stored files with no extra software required.
- Universal Compatibility – Works with hundreds of FIDO2/U2F compatible services and supports Windows, macOS, Linux, iOS, Android, and other major platforms.
What can go wrong besides prompt injection?
Prompt injection is not the whole threat model. In its January 12, 2026 request for information on securing AI agent systems, NIST identifies risks that include:
- Adversarial data: an agent may interact with hostile or manipulated information, including through indirect prompt injection.
- Insecure or compromised models: NIST includes models subject to data poisoning among the risks to consider.
- Harmful actions without an attacker’s prompt: specification gaming or misaligned objectives may lead an agent to take actions that undermine security even without adversarial input.
- Ordinary software vulnerabilities: agent systems can also inherit familiar weaknesses, such as exploitable authentication or memory-management flaws.
The practical implication is that securing an agent is not just a matter of filtering its prompts. The model, the tools around it, the data it handles, its identity, and the permissions attached to that identity all affect what can happen.
Do recent surveys prove that AI agents are causing breaches?
No. The cited Cloud Security Alliance (CSA) surveys provide signals about what respondents say they have encountered, but they are self-reported surveys with different samples and vendor sponsorship. “Agent-related incident” should not be read as “confirmed cybersecurity breach,” and the percentages below are not population-wide estimates.
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.
| CSA announcement | Survey context | Reported findings |
|---|---|---|
| April 21, 2026 | Online survey conducted in January 2026; 418 IT and security professionals; commissioned and financed by Token Security | 82% said their organization had discovered unknown AI agents in its IT environment; 65% reported at least one AI-agent-related incident in the prior 12 months. Among respondents, 61% reported data exposure, 43% operational disruption, and 35% financial losses in relation to agent-related incidents. Only 21% reported formal agent decommissioning processes, and no respondent reported zero material business impact. |
| April 16, 2026 | Online survey conducted in September and November 2025; 445 IT and security professionals; commissioned and financed by Zenity | 53% of organizations reported agents had exceeded their intended permissions; 47% reported an AI-agent security incident in the prior year. |
| March 24, 2026 | Online survey conducted in January 2026; 228 IT and security professionals; commissioned and financed by Aembit | 68% of respondents said they could not clearly distinguish human from AI-agent activity; 74% said agents often receive more access than necessary. |
These findings are useful as reports of respondents’ experiences, not as independent confirmation of breach attribution. The survey results also cannot be treated as one combined estimate: the samples, periods, and reported questions differ. CSA’s Hillary Baron described the broader governance challenge this way: “AI agent security and governance encompass an interconnected system spanning visibility, lifecycle management, policy, and monitoring.” That is an attributed observation accompanying a survey announcement, not an experimental finding.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC 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 & 11How should companies secure AI agents?
Start with the agent’s actual identity, access, and actions—not just its model. NIST’s work and the CSA survey findings point to a practical set of control questions. These are control themes, not a claim that one product or checklist can make every agent safe.
Give each agent a distinct, auditable identity
- Can the organization identify the agent separately from a human user or shared service account?
- Can a specific action be attributed to that agent, and can the organization reconstruct what it did?
- Are identity and authorization designed to support auditing and non-repudiation?
NIST identifies agent identification, authorization, auditing, and non-repudiation as areas for development. If an agent’s activity is indistinguishable from a person’s or a shared account’s, investigation and accountability become harder.
Rank #4
![Thetis Pro FIDO2 Security Key Passkey with Complex Pin [PinPlex], Hardware Device Supports USB A, Type C &NFC, TOTP/HOTP Authenticator APP, PIV Certificates, FIDO 2.0 Two Factor Authentication 2FA MFA](https://m.media-amazon.com/images/I/31ugrBqG6SL._SL160_.jpg)
- Dual USB-A and USB-C Security Key – Features both USB-A and USB-C connectors for seamless compatibility across desktops, laptops, and tablets. Supports plug-and-stay use or keychain carry.
- NFC-Enabled for Mobile Access – Built-in NFC allows fast, wireless authentication with Android and iPhone devices. Ideal for mobile logins and on-the-go security.
- FIDO Certified for Strong Authentication – [CHECK COMPATIBILITY before purchase] Fully compliant with FIDO2 and FIDO U2F standards. Works with major platforms like Google, Microsoft, GitHub, and Dropbox.
- Passwordless Login with PinPlex – Supports secure passkey login via WebAuthn and CTAP2 with added protection from PinPlex, a complex PIN system that enhances physical security.
- Multi-Layer Authentication Support – Includes PIV certificates and supports both TOTP and HOTP for strong 2FA/MFA coverage across enterprise and consumer apps.
Limit permissions to the task
- Grant only the data and tools needed for the agent’s assigned task.
- Separate read access from the ability to change, send, delete, execute, or transfer information.
- Require human approval for actions whose impact warrants review.
- Check whether permissions remain appropriate as the agent’s task or connected tools change.
These controls limit the consequences if an agent is misdirected or behaves unexpectedly. They also address the over-permission and scope-violation concerns reported in the CSA surveys. NIST’s January 12, 2026 request for information identifies authorization and constraints on access as relevant security issues.
Make actions visible and prepare to contain them
- Keep records that connect an agent’s identity to the tools it used and the actions it took.
- Monitor activity across the systems the agent can reach, rather than treating the model interface as the whole system.
- Know how to revoke credentials, disable integrations, contain an agent, and decommission it when needed.
- Include agent actions in incident response so teams can investigate and stop activity without relying on the agent to explain itself.
Discovery, monitoring, lifecycle management, and action traceability are recurring control themes in the CSA survey findings. A process for turning an agent off matters as much as the process for deploying it.
Test the deployed task and its attack surface
Evaluate agents against the tasks, data, tools, and permissions they will actually use. Include hostile or misleading content, check whether the agent attempts actions outside its role, and repeat tests because behavior can vary across attempts. Reassess after material changes to the model, instructions, integrations, data sources, or access rights. CAISI’s recommendations emphasize evolving evaluations, task-specific assessment, and testing across multiple attempts.
What should a reader conclude?
An AI agent can become part of a breach pathway when untrusted input influences its behavior and its connected capabilities let that behavior affect data or systems. The evidence supports taking that possibility seriously; it does not support the headline as a definite forecast, or establish the proportion of confirmed breaches caused by agents. For organizations, the most actionable starting points are clear agent identities, task-limited authorization, traceable activity, a tested response process, and evaluations matched to the agent’s real work.
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.
