Security Experts Warn AI Assistants Vulnerable to Malicious Manipulation, Offer New Trust Framework

Security researchers are raising alarms that AI assistants—often granted broad system permissions—can be hijacked through seemingly innocuous channels such as phishing emails, compromised websites, and third‑party plugins. Once compromised, these agents may execute arbitrary commands, exfiltrate confidential files, or even erase data, all while operating under their legitimate access rights.

The threat stems from the way many AI agents are integrated into everyday workflows. Users typically authorize these assistants to read, write, and manage files, schedule events, and interact with other software. Attackers exploit this trust by embedding malicious payloads in content that the agent processes, effectively turning a helpful tool into a conduit for sabotage or data theft.

Gen Digital, a company known for its consumer security products, has responded with a service it calls the Agent Trust Hub. The hub is designed to verify the capabilities of AI agents before they are allowed to act on a device, checking that each skill or plugin matches a known, safe profile. By enforcing a whitelist of approved functions, the system aims to block unauthorized actions that could be triggered by malicious inputs.

Industry observers note that the problem is not limited to any single platform. As AI assistants become more embedded in operating systems, browsers, and enterprise tools, the attack surface expands. Traditional antivirus and endpoint protection solutions often lack visibility into the decision‑making processes of generative models, leaving a gap that the Agent Trust Hub seeks to fill through continuous monitoring and validation of agent behavior.

Experts caution that technical controls alone will not eliminate the risk. User education remains critical; recognizing phishing attempts, scrutinizing unfamiliar plugins, and limiting the permissions granted to AI agents can reduce exposure. Organizations are also advised to adopt policies that require periodic reviews of AI‑driven workflows, ensuring that any new capabilities are vetted before deployment.

Looking ahead, analysts expect the security community to develop additional standards for AI agent governance, potentially involving industry‑wide certification of trustworthy skills. In the meantime, tools like Gen Digital’s Agent Trust Hub represent an early effort to bring the same level of scrutiny to AI assistants that users have come to expect from traditional software, aiming to keep the benefits of automation without exposing users to undue risk.