Your AI Bodyguard Has Arrived

If your company handles sensitive data, it’s time to treat AI like a high-risk meeting room, not a public chat forum.

Marv is a company-wide AI privacy tool that intercepts and anonymises prompts before they reach generative AI tools like ChatGPT. Installed at the IT level, it offers automatic, frictionless protection across all company devices. It also serves as an AI training lens, helping firms evaluate how well teams use prompting and AI tools.

The discreet, seamless browser tool protecting Asia’s legal, finance, and enterprise teams

When ChatGPT swept into public view in late 2022, it promised to turn typing into a superpower. But in boardrooms across Asia, one uncomfortable reality quickly surfaced: every typed prompt could be a data leak waiting to happen.

That’s the problem MagicMirror set out to solve with Marv, its quietly powerful browser extension described by co-founder Daphna Wegner as "a silent bodyguard for your data."

The Perils of Prompting Without Protection

Imagine an associate at a Singaporean law firm feeding a sensitive client contract into ChatGPT to produce a quick summary. Unwittingly, she’s just submitted names, financial clauses, or case identifiers to OpenAI’s servers. Without safeguards, such data could be stored, analysed, or potentially breached.

Marv steps in before that happens. Installed as a browser extension across employee devices, it scans each AI prompt in real time, scrubs it of sensitive data, submits an anonymised version to the AI model, and then reinserts the correct information once a response is received.

The result? Your team still gets the speed and power of AI, without ever exposing confidential information to the cloud. You might also be interested in how AI is recalibrating the value of data in today's business landscape [/business/ai-recalibrated-value-of-data].

"Our application reviews data for sensitive content. If any exists, it is anonymised before sending. The process is seamless," says Daphna Wegner, Co-Founder, MagicMirror.

"Our application reviews data for sensitive content. If any exists, it is anonymised before sending. The process is seamless," says Daphna Wegner, Co-Founder, MagicMirror.

Enterprise-Grade Simplicity

Unlike consumer tools, Marv doesn’t rely on staff remembering to activate a plugin or toggle a setting. It is force-installed at the IT level, which means there’s no opting out. Every browser session is protected, every time.

"It’s very easy to manage," says Wegner. "You cannot forget to switch it on and off. Once it’s up and running, it’s simply there, working in the background."

"It’s very easy to manage," says Wegner. "You cannot forget to switch it on and off. Once it’s up and running, it’s simply there, working in the background."

Centralised admin controls let company leaders establish rules, permissions, and restrictions by department or role. Legal teams can mask client identifiers; HR departments might shield salary data; finance teams can flag account numbers. Meanwhile, audit trails and analytics give transparency on what’s being redacted, sent, or blocked across the business. This kind of robust data handling is crucial, especially as companies navigate the complexities of AI ethics and governance, like those being discussed in India's AI Future: New Ethics Boards.

The Tool That Also Teaches

While security remains Marv’s first mission, it does something quietly clever in the background. By capturing and analysing anonymised user behaviour across the firm, it builds a picture of how well teams are using AI.

Prompts, patterns, and performance are revealed. Which departments are thriving with AI assistance? Which are floundering with vague queries? Who needs training? Who might be ahead of the curve?

"It’s a kind of training assessment," says Wegner. "It reveals which people or departments are best utilising AI and which can be improved."

"It’s a kind of training assessment," says Wegner. "It reveals which people or departments are best utilising AI and which can be improved."

A Growing Class of AI Defenders

MagicMirror isn’t alone in this new space. HiddenLayer, Preamble, and Lakera have all emerged with similar propositions aimed at preventing AI misuse, poisoning, or data leakage. But few combine real-time filtering with behavioural insights quite as neatly as Marv. The rise of such tools highlights a broader trend in the tech industry, where AI Browsers Under Threat as Researchers Expose Deep Flaws. For a deeper dive into the technical challenges and solutions in AI security, you can explore resources from the AI Security Alliance, an organisation dedicated to securing AI.

In Japan, the Kaizen philosophy of continuous improvement might best describe what this breed of tools is enabling: a company-wide system of AI guardrails that not only protects but improves. The AI bodyguard isn’t just catching threats; it’s also shining a light on how your people are growing with the technology in hand.

A Culture of Safe Experimentation

The subtext of MagicMirror’s approach is refreshing. It doesn’t block AI outright, nor does it add friction to its use. Instead, it encourages safe exploration. Employees can prompt with confidence, secure in the knowledge that sensitive data is being guarded invisibly.

"Through the application of technology, it’s increasingly possible to both safeguard and improve ourselves," adds Wegner. "But only if we accept a mentality of proactivity."

"Through the application of technology, it’s increasingly possible to both safeguard and improve ourselves," adds Wegner. "But only if we accept a mentality of proactivity."

For companies handling personal data, legal documents, medical records, or confidential IP, this balance is critical. Productivity should never come at the cost of privacy.