
Is Psychological AI safe on LinkedIn in 2026?
Risky — LinkedIn fingerprinted Psychological AI (we checked it ), and Psychological AI has since been removed from the Chrome Web Store (Nov 2026 snapshot). If an old copy of Psychological AI is still installed, LinkedIn keeps detecting it on every visit. Our automated scan of Psychological AI 1.0.71 found code that references LinkedIn's session cookie li_at (11 matches in 1 files); whether Psychological AI sends that cookie anywhere needs a manual dataflow review.
Updated · LinkedIn AED snapshot Nov 2026
Does LinkedIn detect Psychological AI? What we found in its code
LinkedIn audited this extension and fingerprinted its files in its client code. Since then, the extension has been removed or delisted from the Chrome Web Store (either by Google for policy violations, or by the developers to avoid bans).
Scanning risk: If you currently have an old package of this extension active on your drive, LinkedIn continues to scan your browser using HTTP requests to its unique internal files.
What our automated scan found in Psychological AI's code
We downloaded Psychological AI version 1.0.71 from the Chrome Web Store on and searched its 16 code files for the patterns below. Each line shows the first place we found it.
- Psychological AI's code references
li_at— LinkedIn's session cookie, the one that keeps you logged in. 11 matches in 1 file — first at scripts/background.js:72. - Psychological AI's code references
JSESSIONID— LinkedIn's CSRF cookie, needed to call LinkedIn's internal API as you. 11 matches in 1 file — first at scripts/background.js:73. - Psychological AI uses the Chrome cookies API (
chrome.cookies), which reads cookies of every site it has access to. 11 matches in 1 file — first at scripts/background.js:71. - Psychological AI reads or writes
document.cookieinside pages. 2 matches in 1 file — first at main.js:2. - Psychological AI creates synthetic clicks or keystrokes (
dispatchEvent/MouseEvent) — LinkedIn can tell these from real input (isTrusted=false). 5 matches in 1 file — first at main.js:2. - Psychological AI runs code built at runtime (
eval/new Function). 2 matches in 1 file — first at main.js:2. - Psychological AI makes its own network requests (
fetch/ XHR / WebSocket). 4 matches in 3 files — first at main.js:2. - Psychological AI's code mentions linkedin.com. 13 matches in 2 files — first at scripts/background.js:71.
What this scan is — and isn't. It is an automated static search of the shipped package, not a full teardown: a match means the pattern is in the code, not that Psychological AI misuses it, and no match doesn't prove absence (minified or runtime-built code can hide it). We haven't traced where Psychological AI sends data. How we test
How LinkedIn checks for Psychological AI: the AED probe
That verdict isn't a guess. This is what LinkedIn's own page code does on every visit — simplified from its production bundle, which the BrowserGate investigation reverse-engineered line by line:
// chrome-extension://hegppmjkecopkbicfkecnlemdincdnfj/images/card/info.svg

How LinkedIn fingerprints your browser while you use Psychological AI
Even after a cookie is copied out, LinkedIn keeps one more receipt on every request that follows — a 48-point snapshot of the exact machine making it. Faking it convincingly means reproducing that machine's quirks, not just editing a string.
Click any signal above for what it means. These are illustrative values — this demo never reads your actual device.
Canvas render hash: Draws a hidden shape and hashes the pixels — GPU/driver quirks make it near-unique. (illustrative — e.g. a1f9…3e0c)
WebGL parameters: 65+ GPU/renderer values read straight from the graphics driver. (illustrative — e.g. ANGLE (Apple, M2 Pro))
AudioContext: Plays an inaudible tone and hashes tiny hardware-specific waveform differences. (illustrative — e.g. 124.0432…)
Installed fonts: Measures text rendering to infer which fonts are installed on your machine. (illustrative — e.g. 38 fonts)
Local IP (WebRTC): A WebRTC leak reveals your real local network IP, even behind a VPN. (illustrative — e.g. 192.168.1.x)
Timezone ×2: Checked two different ways — a mismatch between them is itself a signal. (illustrative — e.g. Europe/Warsaw)
Hardware & OS: CPU core count, device memory, screen size/depth, platform string. (illustrative — e.g. 8 cores · macOS)
navigator.webdriver: Chrome's own automation flag — true when a page is driven by Selenium/Puppeteer. (illustrative — e.g. false)
Cameras & mics: Enumerates connected media devices without asking permission to use them. (illustrative — e.g. 1 camera · 2 mics)
Incognito flag: A filesystem-quota heuristic that detects private-browsing mode. (illustrative — e.g. false)
This extension runs inside your real browser, so its requests always carry your real fingerprint — that isn't something it controls, it's simply where its code executes.
Similar LinkedIn tools we checked
Same architecture and risk as Psychological AI — each with its own code audit.
- Is Spark Hire Recruit Sourcing for Chrome safe on LinkedIn?MEDIUM
- Is PhantomBuster safe on LinkedIn?HIGH
- Is InTouchApp Phone Contacts & Data Saver safe on LinkedIn?MEDIUM
- Is Linked Assist safe on LinkedIn?MEDIUM
- Is XING Transfer to LinkedIn safe on LinkedIn?MEDIUM
- Is LinkedIn Email Finder & Phone Number Finder by LeadStal safe on LinkedIn?MEDIUM