Introduction
For years, the default image of internet censorship was a blacklist: authorities named forbidden domains or IP addresses, and everything else stayed online. Circumvention became a game of finding gaps in the list. That model is fading. Across several countries, network control is shifting to an allowlist logic—only approved services remain reachable, and everything else is dark by default. This article explains how that shift works, why familiar tools like VPNs and Tor are losing ground, and what a new architecture built around local agents, sharded relays, and lightweight coordination might offer.
What Happened
Measurement reports and news coverage from 2025 and 2026 show the change in motion. In Russia, mobile internet shutdowns became routine, with only a short list of approved services staying online; one monitoring report recorded shutdowns in 54 regions on September 27, 2025, with active whitelists in 30 of them. Iran experienced a nationwide shutdown beginning January 8, 2026, followed by another from February 28 that monitors described as the longest nationwide blackout on record, dropping connectivity below one percent of normal before a partial restoration in late May. Meanwhile, leaked documents from a Chinese Great Firewall vendor indicated that similar technology has been deployed outside China, including in Ethiopia, Pakistan, Myanmar, and Kazakhstan.
Four Stages of Control
The article frames modern censorship as four overlapping eras. First, everything works except listed targets, enforced through DNS poisoning and IP blocks. Second, deep packet inspection classifies traffic and blocks anything that looks suspicious, including encrypted protocols. Third, nothing works except listed targets, seen in Russian mobile whitelists and Iran's tiered access during shutdowns. Fourth, a domestic intranet becomes the default while the global internet is treated as an exception, as in North Korea's Kwangmyong network.
Why Current Tools Are Straining
Existing circumvention tools such as Tor with Snowflake, Psiphon, Lantern, and various VPN protocols still serve millions, but they face pressures no single tool handles well. Blocking differs by internet service provider and changes over time. When only approved destinations are reachable, tunneling through allowed IP ranges works until censors demand that cloud providers separate whitelisted address pools. Censors are also beginning to use AI-assisted traffic classification, and satellite links can be jammed or criminalized. Mesh networking helps in dense gatherings but requires preinstalled apps and nearby peers, while blockchain-based naming and storage remain impractical for ordinary users because of wallets, fees, and reachability limits.
Why This Matters
The shift to allowlists changes the threat model for anyone designing censorship-resistant software. A tool that simply hides inside permitted traffic can be neutralized once the censor identifies the dependency. Techniques developed against one national system are increasingly sold as commercial products to other countries, so a local fix is rarely a lasting fix. At the same time, users need a one-tap experience without wallets, private keys, or technical expertise. Any viable design must assume that internet service providers act as independent adversaries, that control traffic must be kept tiny, and that no single point of failure can be allowed.
Key Takeaways
- Network control is moving from blocklists to allowlists, where only approved services remain reachable.
- Recent shutdowns in Russia and Iran show whitelists enforced at the infrastructure level, not just the service level.
- Traditional VPNs and anonymity networks struggle with per-ISP blocking, AI-assisted classification, and allowlist-aware countermeasures.
- A proposed design uses local on-device agents, diverse transport pools, and sticky relay shards to limit what any single censor identity can learn.
- A thin coordination layer handles operator identity and reputation without touching the user's data path.
- The architecture explicitly avoids remote inference during connections, keeping models local and control traffic under a strict byte budget.
Conclusion
The allowlist era does not mean the end of open networking, but it does require a different starting point. Instead of one clever disguise, the design sketched here relies on many options chosen by a local agent based on what actually works on each network. Persistent shards limit exposure, while a lightweight ledger coordinates operator reputation without exposing user traffic. The approach remains a hypothesis assembled from published research rather than a finished product, and several challenges—including scaling trust and surviving total physical disconnection—are left unsolved. Still, it offers a direction for building networks that assume restriction is the default and still try to survive it.

Discussion
Join the conversation
Thoughtful reactions, questions, and follow-up ideas help shape the next story.