AI Policy

Last updated: July 26, 2026

Why this page exists

A lot of security products now send your data through an AI model without telling you. We think you should not have to guess. This page says exactly where AI is involved in Intersafe and where it is not.

The summary: no AI touches your browsing. AI helps us build the software. Everything below explains both halves of that, and how the second half is checked.

No AI runs on your browsing

Intersafe's protection contains no artificial intelligence of any kind. There is no AI model inside the extension, and the extension does not call one.

Specifically:

• No page you visit is sent to an AI model or an AI company
• No clipboard content is sent to an AI model
• No browsing history, URL, or form data is sent to an AI model
• No AI decides whether to warn you or block something
• No AI profiles you, scores you, or learns from your behavior

Threat detection works by pattern matching and structural checks that run entirely on your device, plus Google Safe Browsing for known-dangerous sites. There is no model making a judgment call in the middle. The same page produces the same result every time, and the detection logic works with no network connection at all.

Our Privacy Policy covers what data does leave your browser, which is very little and never includes page content.

Where AI is used: building Intersafe

We use AI as a development tool. Intersafe is built with the help of Claude, an AI assistant made by Anthropic. That includes writing and reviewing the extension's code, and writing the detection patterns that recognize attacks such as fake CAPTCHA prompts and clipboard hijacking.

Some of the writing on this website and in the extension is also drafted with AI assistance.

We are telling you this because it is true and because it affects you indirectly, even though no AI runs on your machine. A bad detection pattern can miss a real attack or flag a safe page. So the question that actually matters is not whether AI wrote it. It is how carefully it gets checked before it reaches you.

How the detection logic gets checked

Every change to Intersafe's detection has to pass automated verification before a release is built. None of these checks are AI. They are ordinary deterministic tools, and they fail the build rather than warn.

• Pattern safety linter. Every detection pattern is run through a tool that rejects patterns capable of freezing a browser tab. This class of bug is easy to write by accident and hard to spot by eye. The linter has caught real ones before they shipped.
• Shape tripwire. A separate check fails the build if a known-dangerous pattern shape is introduced anywhere in the codebase, so the problem cannot come back in a new file.
• Test suite. A suite of tests covering detection patterns, scan limits, and feed verification must pass before a release is packaged.
• Structural evidence requirement. Text alone never triggers a block. Intersafe requires real attack machinery on the page, such as a hidden element, a fullscreen overlay, or clipboard manipulation. This is why a news article or a forum thread about these attacks is not flagged.
• Scan limits. Detection runs against capped amounts of content so an unusually large page cannot slow your browser down.
• Signed pattern updates. Detection pattern updates are cryptographically signed, and the extension rejects any update that is not signed with our key. The signing key is held offline and is never stored on our servers, so even if our servers were compromised, nobody could push a malicious pattern to your browser.
• Store review. Every release goes through Chrome Web Store review before it reaches you.

Releases are approved by a person. Nothing ships automatically.

What we are not claiming

Transparency is worth little if it only lists the good parts.

• Intersafe is a small operation. Detection logic is verified by the automated checks above and by testing. It does not get line-by-line review from a large in-house security team.
• Automated checks are not proof. They catch the failure modes we know about. They cannot rule out every mistake.
• False positives happen. A legitimate page can occasionally trip a detection. You can add any site to your allowed domains list.
• False negatives happen. Because detection is pattern based rather than AI based, a genuinely new attack shape may not be recognized until we add a pattern for it. This is the honest trade-off for keeping your browsing private and local.
• Intersafe is one layer. It is not a replacement for care, for keeping your system updated, or for antivirus software.

Telling us we got it wrong

If Intersafe flagged something safe, or missed something dangerous, email support@intersafe.app with the site address and what happened.

Reports about a missed attack go to the front of the queue, because they usually mean a pattern needs updating for everyone. We reply as quickly as we can, and within 5 business days at the latest. If a report turns out to need longer investigation than that, we will tell you it is being worked on rather than leave you waiting without an answer.

If this ever changes

If we ever add AI to the product itself, you will not have to find out by reading this page. Before it ships we will:

• Update this page to say plainly what the AI does, what data it can see, and whether you can turn it off
• Announce it on this website rather than slipping it into a release quietly
• Tell you inside the extension itself, so the notice reaches you where you actually use Intersafe instead of sitting on a page you have no reason to revisit

If such a change also affects what data leaves your browser, the notice period in our Privacy Policy applies as well: account holders get at least 30 days' email notice before it takes effect.

We will not quietly start sending your browsing to a model.

EU AI Act

The EU AI Act (Regulation (EU) 2024/1689) sets obligations for products that place an AI system on the market. The Intersafe extension does not contain or call an AI system, so it does not place one on the market. We publish this page as voluntary transparency rather than because an AI-specific rule requires it.

Contact

Questions about this policy? Email us at support@intersafe.app.