Talos spent a blog post this week explaining something operators have been treating as a future problem. Malware crews are parking command-and-control in public blockchains, using the same ledgers your treasury team already calls immutable. You can seize a VPS. A confirmed transaction does not take a ticket. That is the product. Most cybersecurity programs still write incident response as if the next callback will have a hostname, a registrar, and an abuse desk that picks up.
Same week, a suspected Chinese APT rebuilt EggStreme so it stops making noise. Public board for instructions. Quiet implant on the box. Your two favorite moves, sinkhole and hash-block, fail together and nobody at the vendor is going to fix that with a feed update.
The dead drop doesn’t have a landlord
Classic C2 was a box you could name. Hardcoded domains. Bulletproof hosts that still had a billing relationship with someone you could email. You blocked the IP, you burned the domain, you attached a PDF to a registrar, and you told leadership the campaign was contained. Cisco Talos is describing the next step: dead drops on public chains, where the “server” is a transaction anyone can read and nobody can delete.
The irony writes itself. Attackers used to hide C2. Now they publish it on purpose. The chain is the bulletin board. An implant checks a contract, a memo field, or a tiny transfer, then decodes the next stage. There is no single point of failure because the network is the point. You do not get a takedown window. You get an archive.
If your firewall still thinks threat-protection means dropping last week’s bad IPs, this traffic looks like a user hitting a crypto RPC. Plenty of shops already allow that path because a developer, a finance bot, or a leftover chain experiment needed it last quarter. Nobody revoked the exception. The allow list aged into a standing dead-drop client.

Defense in depth still matters here. It has to land on egress, not on a domain your intel vendor will “add by Friday.” Blocking two explorer hostnames is theater if the implant can speak JSON-RPC to any public node, any paid gateway, or a node the operator spun up that morning. Treat blockchain APIs as high-risk destinations, same class as anonymous file lockers. Stop treating them as hobby traffic you will get to later.
This is a bad look for teams that sold leadership a takedown narrative. You cannot sinkhole a public chain. You cannot “request removal” from a mempool. You can watch who on your network is reading the board. That is a smaller, uglier, more honest job than the press-release version of threat intel, and it is the one that still works.
EggStreme went quiet because hashes are cheap
While Talos was mapping C2 you cannot unplug, Bitdefender was watching operators rebuild EggStreme so it stops tripping the obvious sensors. The target set is still government organizations in the Philippines. The toolkit is not. When researchers publish your toys, the cheap response is a recompile. A file hash changes. A domain costs a few dollars. Yesterday’s SHA-256 is a souvenir.
Silent is the design. You trained analysts on beacons, brute-force spray, and loaders that trip every default rule. These operators are betting you will keep waiting for that soundtrack. Fewer callbacks. More living off what is already on the box. A thinner gift to the EDR timeline. Cyber security teams that close the ticket when the hash stops matching are filing the attacker’s change request for them.

Put both stories on one whiteboard and the picture is rude. The channel is public and durable. The implant is quiet and disposable. Sinkholing C2 and blocking hashes fail in the same week for different reasons. Rotation is cheaper than your detection-engineering sprint. Immutability is cheaper than your legal letter to a registrar that no longer sits in the path.
Government networks are the proving ground because they still run mixed estates: old jump boxes, “temporary” internet for a vendor, workstations that can reach anything answering on 443. A quiet loader plus a chain-based dead drop does not need your user to click twice. It needs one machine that can read a public API and a process that does not look like a reverse shell from 2018.
Your cybersecurity stack is still hunting hostnames
Walk your last tabletop. Someone still says we block the C2 domain at the edge. Someone else dumps hashes into the EDR and calls it coverage. Both are fine when the adversary is sloppy. They stall when the dead drop is a public ledger and the implant is a fresh build by Thursday.
The real problem here is muscle memory. You staffed for domain intel, registrar letters, and IOC pipelines. Those pipelines still catch the noisy majority, the commodity loaders, the reused panels. They miss a contract read from a workstation allowed to talk to a public node. They miss an implant that decided to stop chatting.
Keep hunting brute-force noise. It still pays. Do not confuse volume with coverage. The campaigns that will ruin a quarter never spray, never shout, and never give you a hostname to argue about on the bridge.
If your only plan is taking down their server, you do not have a plan.
Ask a sharper question in the war room: which of our hosts can fetch arbitrary data from a public chain, and which processes on those hosts have any business doing it? That question survives EggStreme’s next rebuild. A blocklist of last week’s contracts does not.
Do this before the next dead drop, not after
Start with the path, not the actor name. Today, pull egress logs for JSON-RPC patterns, default node ports, and known public gateway hosts. Flag any server, jump box, or user subnet that talks to those destinations without a written owner. If finance or a builder legitimately needs a chain API, pin it to a named account, a named host, and a named vendor endpoint. Everything else is an unlicensed C2 client waiting for instructions.
On the host side, stop treating file hashes as the EggStreme problem. Hunt for new persistence, odd parent/child pairs, and processes that take a little network and then go quiet. Memory and behavior beat yesterday’s sample. If you cannot do that on government-adjacent boxes this week, isolate those boxes from general internet RPC until you can. Security hardening here is subtraction. Remove the ability to fetch a dead drop, then argue about detection.
For incident response, rewrite the containment step that currently says sinkhole and block indicators. Cut egress to public chain interfaces. Rotate local credentials. Assume the implant can be rebuilt before your IOC feed refreshes. Practice that on a tabletop where the only “C2” is a transaction ID. If the room stalls, you found the gap.
Keep doing this. Revisit the RPC allow list every sprint. Watch for new developer tools that speak to chains by default. Recheck that your threat detection actually alerts on low-and-slow outbound to those APIs, not just on malware family names. The next silent variant will not wait for your intel vendor to invent a new noun.
Sources
- The challenge of defending against bulletproof dead drops on public blockchains
- New EggStreme Malware Variant Goes Silent to Evade Security Tools in Government Attacks
Take Control of Your Server Security
Don't let brute-force attacks slow you down. Try IPBan Pro risk-free for 30 days.
Secure. Automated. Lightweight.
Take Control of Your Server Security
Don't let brute-force attacks slow you down. Try IPBan Pro risk-free for 30 days.
Secure. Automated. Lightweight.
