Bitdefender launches AI Guardian beta for Mac agents

Bitdefender has launched AI Guardian, a security tool for autonomous AI agents, as a free public beta for macOS.

The software is aimed at developers, technical practitioners, and other users who rely on AI agents to access tools, files, and credentials on their behalf. It is designed to give users more oversight of what those agents can do and to stop unsafe actions before they happen.

The launch targets a growing area of cybersecurity as AI agents move beyond text generation into tasks that require direct access to systems and sensitive data. That shift has raised concerns that agents can be manipulated through prompt injection, poisoned tools, or exposed credentials.

Bitdefender cited independent research that tested 20 leading AI agents against more than 1,300 tool-poisoning attempts and found an average attack success rate of 36.5%. One model in the research was manipulated 72.8% of the time.

Separate analysis the company referenced found more than 1.2 million exposed AI service secrets in 2025, up 81% year on year. That total included more than 24,000 credentials leaked through public Model Context Protocol configurations.

How it works

AI Guardian runs as a background service on macOS and connects to supported agent environments through dedicated integrations. The initial version supports Claude Code 2.1.121 or later and OpenClaw 2026.6.6 or later.

The product uses a three-stage model: it sets a policy baseline for permitted tools, files, and actions; checks each attempted action against that baseline in real time; and returns a verdict before the action proceeds. Those verdicts are allow, flag, or block, and each decision is recorded in an auditable log.

Prompt analysis takes place on the device, meaning prompt text does not leave the machine. Some checks, including URL reputation, use Bitdefender's cloud services.

According to Bitdefender, AI Guardian can detect and block prompt injection attempts, inspect Model Context Protocol tools, validate agent skills before execution, and control access to exposed application programming interface keys, secure shell keys, and system credentials.

Broader push

AI Guardian is the third product Bitdefender has introduced this year for what it describes as the agentic ecosystem. The other two are Agent Skill Scanner and VPN for AI Agents.

Together, the products target three areas of AI agent risk: what software or skills they install, how they connect to external services, and what actions they take once operating inside a user environment. The latest launch adds a consumer-facing offering to that effort, though it is built primarily for technical users rather than mainstream consumers.

The release also reflects a broader shift in security thinking, with suppliers starting to treat AI agents as actors that need their own controls rather than simply as features within a larger application. In practice, that means setting policies for agent behaviour much as companies already do for employees, devices, and workloads.

The initial release is available in English only. Bitdefender plans to add support for other operating systems but has not provided a timeline.

"AI agents are becoming a direct extension of the users who rely on them, inheriting the same security risks that come with that role," said Ciprian Istrate, senior vice president of operations, Consumer Solutions Group at Bitdefender. "This rapid shift means security can no longer stop at protecting the human alone; the agent itself has become its own entity to secure. Bitdefender built AI Guardian in response, giving developers and other technical users the confidence to let their agents code, build, and execute tasks securely."