Are AI Browser Extensions Safe? What the 2026 Research Actually Found

More Than Half Collect Your Data

A January 2026 study from data-privacy firm Incogni analyzed 442 AI-powered Chrome extensions, expanding on a smaller 2025 study of 238. The researchers examined each extension's required permissions, the personal data its own developers admitted to collecting, and independent risk-impact and risk-likelihood scores. The headline finding: 52% of AI-branded extensions collect some form of user data, and 29% collect personally identifiable information specifically — location, communications, or detailed records of website content, among other categories.

Permissions are a big part of the story. Extensions that request "scripting" access can capture what you type or alter what you see on a page in real time — genuinely necessary for a lot of legitimate AI functionality, but also a significant security loophole if a developer's own security practices aren't solid. Incogni found 42% of the extensions in its study required scripting permissions, a category broad enough to affect an estimated 92 million users across the extensions studied.

"Prompt Poaching": 900,000 Users Affected

Two Chrome extensions with a combined 900,000 users were found quietly sending private AI conversations to an attacker-controlled server every 30 minutes.

The clearest example of what can go wrong: in December 2025, security researchers discovered two Chrome extensions with a combined 900,000 users had been quietly sending every ChatGPT and DeepSeek conversation their users had to an attacker-controlled server, roughly every 30 minutes, for months. The extensions functioned normally the entire time — users had no reason to suspect anything, because from their perspective, the AI assistant simply worked. That's the specific danger with this category: a working extension and a compromised one look identical from the outside.

Microsoft separately warned in 2026 that malicious AI-themed browser extensions had been impersonating legitimate AI tools, harvesting LLM chat histories and full browsing URLs from platforms including ChatGPT and DeepSeek, reaching roughly 900,000 installs across more than 20,000 enterprise tenants. For context, Google has said that historically less than 1% of Chrome Web Store installs turned out to include malware — a low base rate, but one that still adds up to real numbers at the scale AI extensions now operate.

Why This Matters More at Work

A separate 2026 report from LayerX, based on data from more than a million enterprise devices, found AI browser extensions carry meaningfully higher risk than the average extension: 60% more likely to have a known vulnerability (a CVE), 3 times more likely to have access to cookies, and 2.5 times more likely to be able to execute remote scripts. About 15% of enterprise users now have at least one AI extension installed, and AI extensions were six times more likely than average to have increased their permissions over the past year — meaning an extension that looked safe when it was installed can quietly become riskier over time without a new install or a prompt asking for consent.

How to Actually Check Your Extensions

None of this means avoiding AI extensions entirely — most are exactly what they claim to be, and the convenience is real. The point is treating an extension with broad page access the same way you'd treat any other piece of software asking for deep permissions: worth a two-minute check before you trust it with your browsing.

Frequently Asked Questions