Why Your VPN Isn't Enough: A Deep Dive into Online Anonymity in the Age of AI

Introduction: The End of the "Padlock in the Browser" Era

Just a decade ago, the green padlock next to a website address gave us a comforting sense of security. "It's HTTPS, nobody can see my passwords, I'm safe," we thought. Today, encrypting the content itself (the payload) is standard practice (just as it is in encrypted messaging apps), but the fight for privacy has moved to an entirely different level: metadata and behavioral analysis.

The question is no longer whether someone can "eavesdrop" on your conversation. Algorithms already know who you talk to, for how long, how often, and even what emotional state you're in. They learn all of this without breaking a single bit of the encryption protecting the content. Join me for an in-depth look at a reality in which incognito mode is a joke and a $5-a-month VPN can be the final nail in the coffin of your anonymity.

Part I: A Network That Sees More Than You Think

The "Envelope" Problem: SNI and Encrypted Client Hello (ECH)

Let's start with the basics. Imagine you're mailing a letter. The letter itself is encrypted and unreadable, but the recipient's address is written on the envelope in big block capitals. For years, that's exactly how HTTPS worked. Because of SNI (Server Name Indication), your internet service provider (ISP) or your company's network administrator couldn't see what you were reading on a website, but they knew exactly which website you were visiting. They knew you connected to a dating site at 8:00 p.m. and to a site about sexually transmitted diseases at 11:00 p.m. The content was secret. The context was public.

In 2025, Encrypted Client Hello (ECH) finally started to see wider adoption. To stick with our analogy, this technology puts one envelope inside another. The outer envelope is addressed to a generic provider (such as Cloudflare), and only the inner, encrypted envelope contains the actual destination address.

Authoritarian regimes and censorship systems hate ECH. Since they can't block specific websites hidden behind ECH, they start blocking entire IP ranges belonging to cloud providers, or the protocol itself. On top of that, ECH adoption is slow. Giants like Cloudflare and Fastly are on board, but traditional hosting providers still leave us "exposed" online, with our browsing habits on full public display.

RevealNet: When the Infrastructure Becomes the Spy

Until recently, we believed that correlating a user's traffic (for example, linking a connection entering the Tor network with one leaving it) required a powerful, centralized supercomputer collecting logs from around the world.

Meet RevealNet. Research published in 2025 shows that this process has been decentralized. Modern programmable network switches (built on the P4 standard) are no longer just "dumb" packet-forwarding devices. They have become intelligent sensors.

These devices generate so-called "flow sketches": compact, cryptographic summaries of your traffic (timing and packet sizes). They exchange these summaries with each other in real time. This means the internet's backbone infrastructure itself can trace a packet's path across many hops. For users of anonymity networks, that's a nightmare. Your VPN provider and the server you're connecting to may be on opposite sides of the world, but the intelligent network in between can "see" that the two data streams match.

Protocol / Tool	ISP Visibility (Domains)	ISP Visibility (Full URLs)	Content Visibility (Payload)	Metadata Visibility (Size, Timing)	Exposure to SNI Analysis
HTTP	Full	Full	Full	Full	N/A (Plaintext)
HTTPS (Standard)	Visible (via SNI/DNS)	Hidden	Hidden (Encrypted)	Full	High (SNI leaks)
HTTPS + ECH	Hidden (Encrypted)	Hidden	Hidden	Full	Low (Depends on destination IP)
VPN	Hidden (VPN IP visible)	Hidden	Hidden	Full (Tunneled)	None (Only the VPN handshake is visible)
Tor	Hidden (Entry Node visible)	Hidden	Hidden	Obfuscated (fixed-size cells)	None

Part II: AI Enters the Game. Your Packets Have an "Accent"

If you thought ChatGPT was the cutting edge of AI, take a look at network traffic analysis. Encryption has become ubiquitous, so "Big Brother" has stopped trying to break it. Instead, it has started analyzing side channels (side-channel analysis).

Website Fingerprinting and Graph Neural Networks

Every website loads in its own distinctive way. An image here, a script there, a font a moment later. This creates a unique rhythm of data transmission, a kind of "fingerprint" for the site.

A new generation of attacks, known as STC-WF (Spatio-Temporal Correlation Website Fingerprinting), uses Graph Neural Networks (GNNs). These algorithms don't treat network traffic as a simple timeline. Instead, they build complex spatio-temporal graphs.

• The spatial dimension: analyzes how different resources load in parallel.
• The temporal dimension: examines how those relationships change within fractions of a second.

The result? An accuracy of 96–98% in identifying which website you're visiting, even if you're using Tor or a VPN. Worse still, Large Language Models (LLMs) are now joining the fight. Researchers have learned to treat sequences of network packets like words in a sentence. The LLM "reads" your encrypted traffic and understands context that was invisible to classical algorithms. This is the end of anonymity as we knew it.

The "Tor over VPN" Myth

Many of us (myself included, I admit) have used this setup: "First I'll turn on a VPN to hide the fact that I'm using Tor from my ISP, and then I'll launch Tor."

The latest research, however, leaves this strategy in tatters. Tor sends data in very specific packages (fixed-size cells of 512 bytes). This creates a rhythm that is hard to conceal. Even if we wrap it in a VPN tunnel (encryption inside encryption), algorithms based on Convolutional Neural Networks (CNNs) can "see" the texture of Tor traffic inside the VPN tunnel.

Detection accuracy? Over 93%, and sometimes as high as 99%. For a state censor, telling apart someone watching Netflix over a VPN from someone sending secret documents over Tor-over-VPN is trivial. It's an illusion of security.

Part III: The Digital Fortress – Operating Systems

If the network is compromised, we have to fall back to defending the endpoint. Antivirus software is useless against zero-day attacks. The only way forward is isolation (Security by Isolation).

Read more: about zero-day attacks

Qubes OS: The Digital Bunker

Qubes OS deserves special attention. It's an operating system built on the assumption that "you will be hacked; it's only a matter of time." That's why Qubes isn't a single system. It's a manager for many virtual machines.

Update: Tor rolls out new CGO encryption – See how the latest Counter Galois Onion update eliminates tagging attacks and strengthens the security of the anonymity network.

Your browser lives in one "box" (a virtual machine, or VM). Your work lives in another. Your passwords live in a third (the Vault). If you open an infected PDF in your "work" box, the malware is trapped inside. It has no access to the passwords in your Vault or to the host system.

The key here is the qrexec mechanism, which its authors describe as a "software air gap." Imagine you want to copy a file from the "Internet" zone to the "Vault" zone. In Qubes, that doesn't happen directly. The system's administrative domain (dom0) takes over the process, checks the security policy, and only then allows (or denies) the transfer. You can configure the system to act like a diode: files can flow into the Vault, but nothing ever flows out. That's a level of security Windows and macOS simply can't match.

Part IV: The Browser – Your Face on the Internet

The most exposed component of all is the browser. This is where the concept of Resist Fingerprinting (RFP) comes in.

Most of us like to stand out. In the world of privacy, standing out is fatal. If you have an unusual screen resolution, a specific set of fonts, and a rare graphics driver version, you're as unique as a snowflake. And easy to track without a single cookie.

The RFP strategy (used in Tor Browser and Mullvad Browser) is based on lying.

• Your system reports that it's in the UTC time zone, whether you're in Warsaw or New York.
• The browser window has standardized dimensions (letterboxing), which explains the odd white bars around web pages in Tor Browser.
• The browser pretends to be running "the most popular version of Windows," even if you're on Linux.

The goal is to be "indistinguishable in the crowd." However, beware of the privacy paradox: if you enable these features in a regular Firefox installation, where only a handful of people use that configuration, you paradoxically become more unique.

uBlock Origin: Hard Mode

For advanced users of the "clearnet" (the regular internet), we recommend switching to uBlock Origin in Hard Mode. This isn't ordinary ad blocking. It's a firewall for your browser.

In this mode, the default rule is: "block everything that comes from outside (3rd-party)." No tracking scripts, no Facebook frames, no Google resources loaded on other sites. It requires users to learn and be patient, because websites will "break." But clicking the gray "noop" column for specific domains (meaning "stop blocking with dynamic rules, but keep filtering ads") quickly becomes a habit that drastically reduces your exposure to surveillance.

Part V: Money Doesn't Smell, but It Leaves a Trail

The Bitcoin blockchain is a public ledger. Everyone can see everything. Companies like Chainalysis have monetized the deanonymization of crypto users. The answer is Monero (XMR).

Monero Under Fire

Monero is a fortress:

1. Ring Signatures: Hide the sender in a crowd of others (decoys).
2. Stealth Addresses: Hide the recipient.
3. RingCT: Hides the amount.

But recent research casts a shadow even over Monero. Attacks targeting the network layer have emerged. Malicious nodes (Sybil nodes) in the Monero network try to map where a transaction physically originates (from which IP address) before it is propagated and obscured (an attack on the Dandelion++ protocol).

There are also reports of a leaked Chainalysis video suggesting the company has methods for tracing Monero, most likely by running its own listening nodes. The authors of these publications recommend that users practice "churning": sending funds to themselves several times at random intervals to break probabilistic links before spending the money.

Part VI: OPSEC – The Human Factor

Finally, the most important point. You can run Qubes OS, connect to Tor over Starlink, and pay in Monero. But if you then log in to your personal Facebook account or use the same nickname you use on a fishing forum, you've lost.

Read more: a short guide to OSINT

Identity Compartmentalization (Personas):

• Public Identity: Banking, government services. Zero darknet.
• Soft Identity: Social media, reviews. Not directly tied to your real name, but still profilable.
• Hard Identity: Full anonymity. Dedicated hardware, cash/XMR only, never used during the same hours as your public identity (timing correlation!).

And don't forget about stylometry. The way you place commas and the words you overuse are your fingerprint. With "Hard" identities, you have to write differently. Or use AI to change the style of your writing.

Conclusion: Between Paranoia and Digital Hygiene

The analysis above paints a picture of a digital battlefield that may seem overwhelming, even dystopian, to the average user. But it's worth pausing for a moment and taking a deep breath. The tools and techniques we've covered, from Qubes OS and ECH to elaborate Monero "churning" procedures, are currently the digital equivalent of a tank. They are powerful, effective, and (for now) provide real, mathematically proven anonymity.

However, you need to ask yourself a key question about your threat model. If you're not the administrator of an onion site, a whistleblower evading intelligence agencies, or a political dissident in an authoritarian country, you don't need to live in a digital bunker 24 hours a day. Using Qubes to look at memes or multi-layer tunneling just to check the weather is a recipe for security fatigue. There's no need for paranoia. Most of us don't need protection against a targeted state-sponsored attack (APT), but rather against the mass, impersonal surveillance of Surveillance Capitalism.

That doesn't mean we can ignore these issues. Quite the opposite. The fact that we have "nothing to hide" in a criminal sense doesn't mean we have nothing to protect.

There are three fundamental reasons to practice good digital hygiene:

1. The right to make mistakes and to change: The internet never forgets. Behavioral data collected today about your mental health, political views, or financial stability could be used against you a decade from now. Algorithms used by insurers, banks, or future employers may judge you based on the "information shadow" you cast today. Anonymity is an insurance policy for your future.
2. Resisting profiling: Even if you never switch to "Hard Mode" with Qubes, adopting the basics (uBlock Origin, a different browser, not publicly broadcasting every aspect of your life) is an act of digital sovereignty. It signals that you refuse to be a free commodity for data brokers.
3. Education and awareness: Technology is moving fast. Understanding how fingerprinting works and how artificial intelligence analyzes metadata lets you use the internet consciously. It helps you understand why "Incognito Mode" doesn't protect you, and make better decisions, even if you sometimes choose convenience over privacy.

The bottom line: you don't have to be invisible to be safe. Awareness is what matters. You don't need to wear a bulletproof vest to pick up groceries, but it's worth knowing when you're entering a monitored area, and you should lock the door to your own home. Think of the tools presented here as a first-aid kit: it's good to know where it is and how to use it, even if we hope we'll never need its most drastic contents.

Aleksander

External sources used in this analysis:

• Navigating Encrypted Client Hello (ECH): Insights from RSAC™ 2025
• RevealNet: Distributed Traffic Correlation for Attack Attribution on Programmable Networks
• Undermining Website Fingerprinting Defenses with Deep Learning
• Qubes OS Documentation: Qrexec
• Friend or Foe? Identifying Anomalous Peers in Monero's P2P Network
• uBlock Origin: Dynamic filtering guide
• OPSEC Guide - Zycher
• Enhancing uBlock Origin with uMatrix
• Bitcoin vs. Monero: Privacy, Transparency, and Transaction Design
• Forensics investigation comparison of privacy-oriented cryptocurrencies
• Leaked Chainalysis Video Suggests Monero Tracking Method
• SaTor: Satellite Routing in Tor to Reduce Latency
• Detecting VPN Traffic through Encapsulated TCP Behavior - Privacy Enhancing Technologies Symposium
• TrafficLLM: Enhancing Large Language Models for Network Traffic Analysis with Generic Traffic Representation
• Two different threat models in website fingerprinting
• A Comprehensive Survey of Website Fingerprinting Attacks and Defenses in Tor: Advances and Open Challenges