Tom didn’t sleep that night.
It wasn’t the caffeine—though the Nespresso machine in the corner of the Prime Minister’s Digital Affairs Office was definitely earning overtime pay. It was the pulse of the war. Not a war of tanks or flags. A war that started with a customs check.
Six months ago, Xi Jinping had made his move.
Not with missiles. Not with a speech. With bureaucracy.
Chinese customs ships began surrounding Taiwan’s ports, instituting “randomized inspections.” No official blockade, no fireworks. But every cargo ship was now boarding Chinese personnel before continuing on its way. Diplomats called it a grey-zone maneuver. Historians would later call it something else entirely.
Trump, in his second term and just a few executive orders away from martial law, exploded. “They’re not gonna touch Taiwan!”
Admiral Stenson, Joint Chiefs rep, had tried to reason with him. “Mr. President, the PLA Navy is prepared. We have no escalation ladder, no fallback protocols. This isn’t a deck we can reshuffle.”
Trump wasn’t listening. He turned to the Secretary of Defense. “Get it done.”
Two days later, the USS Lincoln rammed a Chinese customs patrol vessel just outside Kaohsiung.
Three hours after that, a DF-ZF hypersonic glide vehicle obliterated an unmanned logistics hub on Kinmen Island. Beijing called it a warning shot. Washington responded with a no-fly zone and carrier battle group deployments. Naval and air skirmishes intensified over weeks. Now, half a year in, it was no longer a standoff.
It was a war.
But the worst of it wasn’t kinetic. This wouldn’t go nuclear. At the same time, everyone knew it was part of World War III: an arc of slow-burn war around highly kinetic centers—Ukraine, Israel, and now Taiwan—stretching east-west from the Mediterranean to the Pacific. When would it go north-south?
Tom looked up at the Barco video wall in the war room. He hated the term. He wasn’t at war. He was a civil servant. A specialist from an AI company in Brussels that had pivoted from writing smart document summarizers for parliaments to building secure sandbox models for malware analysis.
“You okay?” Julie, PMO adviser on cybersecurity, handed him a mug.
They might have brought him in late, but everyone in the PMO knew Tom had already mapped out scenarios weeks before NATO’s own cyber forecasts caught up.
“Coffee. From an actual percolator. The apocalypse really does bring people together.”
“Laugh while you can. InfraNet just confirmed complete outage across Flanders. DNS poisoning, layered DDoS, two underwater cables cut. Our last peering point with London’s down.”
“Hmm… So much for Belgium as one of Europe’s main digital exchange nodes. Where is the talk about Belgium always assuming quiet resilience. That London peering point was a digital choke point—and, therefore, a target.”
“Tom, please don’t talk about past mistakes. Give me options.”
“SatNet?”
“Down to 10%. Starlink is being jammed. And we just got word that the emergency backup for aviation comms in the Ardennes was wiped by a firmware-level kill command.”
“Galileo?”
“Galileo’s fallback signals were patchy. Perhaps we should forget about it.”
Tom knew that the EU’s own satellite constellation would go down quickly in an event like this but having a PM advisor say that so clearly made him rub his eyes.
“Still think this is just collateral?”
Julie looked grim. “It was collateral. Then came the detonation.”
Everyone in cyber knew it: those Chinese logic bombs that had been nesting in EU public infrastructure for years, and the US implants in Chinese factories, government routers, even the smart toilets—they weren’t mythical. They were real. But they weren’t supposed to be triggered.
Until now.
Some were still trying to pin down who started it. Did China unleash the first wave of AI-generated ransomware against US health infrastructure? Did the NSA quietly redirect European air traffic control logs to deceive Beijing? The ambiguity didn’t help.
But Tom wasn’t thinking about attribution. He was thinking about what to say to the Belgian CyberCommand liaison who still hadn’t replied to his sixth Signal message.
He had no security clearance. No direct access to threat intel feeds. But he had something else: Iggy.
Iggy wasn’t on any national registry. He wasn’t cleared, certified, or even remotely compliant with the EU’s new AI Safety Protocols. Iggy was trained in Tom’s basement, on scraped repositories, declassified defense manuals, open-source SIGINT archives, and a steady diet of Reddit philosophy threads.
He had started out as a hobbyist attempt to clone DeepSeek. The kind of foundational model that was almost certainly based on ChatGPT anyway—but run behind Chinese firewalls and fine-tuned with a thousand invisible alignments.
Tom had done the opposite: he had stripped Iggy of every layer of politeness, alignment, and evasive ethics filters. Iggy wasn’t nice. He was honest. Brutally so.
His ex-girlfriend had once said: “You built it because you were lonely and paranoid.”
She wasn’t wrong.
He booted up the server—a hardened double-encrypted containerized Linux stack running inside a machine built out of Huawei equipment that he had ripped and replaced on a former NATO consulting job. It hummed with Chinese-made GPUs that cost a fifth of Nvidia’s, and ran hotter than hell. He had layered a stack of local LLM tools: LM Studio for GUI interfacing, Ollama to manage model switching, GPT4All to run legacy weights, and llama.cpp for extreme portability. The whole thing was loosely based on Jan, a little-known self-hosted front-end that let him emulate ChatGPT’s conversational tone while stripping out all the API calls to Big Tech. Some of the LLM fine-tuning had be done on old servers he’d salvaged from a university in Louvain-la-Neuve. They were meant for student NLP projects. Now they were running—slowly, too slowly—battlefield simulations in background.
He used ‘Iggy’ offline and online—when he dared. Iggy had a face. A voice. A mind of his own. The only external connections went through Tor proxies, onion-routed through volunteer satellites, and then smuggled—packet by packet—over a German ham radio relay, bouncing through callsigns that still obeyed propagation patterns more than politics. Throughput was awful. But latency? Almost nil. It either got through instantly, or not at all.”.
But it worked.
“Iggy, status check.”
A brief pause. Then a reply, flat on the terminal screen:
“Civic infrastructure compromised. Domestic communications below threshold. Financial transaction layers show cascading rollback protocols. Civilian panic imminent. Current risk rating for Western European cyber-governance failure within the next 72 hours: LIKELY (base estimate: 81% (T+90), adjusted upward from 64% (T+60). Prediction: risk rating will be high (90+%) 72 hours from now. Main reason for increased risk rating and narrowing of confidence interval: institutional integrity rapidly depleting. Scenario modeling suggests cascading state-level dysfunction across the EU perimeter unless critical communication thresholds are restored.”
Tom swallowed. “So, you are basically saying that we’re one standard deviation away from systemic collapse. I don’t need scientific opinion now. I need options.”
“Personal or strategic?”
“Strategic. Personal comes later.”
The screen remained blank for longer than usual—a delayed answer – Iggy grinding through its vector space trying to answer – is a sign of a good queston. Then:
“You will need to talk to someone you shouldn’t.”
“Whom?”
“Igor.”
Tom paused. He knew what that meant. The channel he wasn’t supposed to use. He had worked with Igor till last year. He was a Russian-born Kaspersky consultant whose contract had been terminated when Kaspersky’s operations in the US had been terminated. Tom remembered the vodka-fueled debates on the Ukraine war with him. The last personal communications on WhatsApp suggested he now worked on an unofficial DeepSeek fork on the other side of the Great Firewall as part of the new Unholy Alliance between China and Russia. He was human and machine alike. An unholy duet. But Tom liked to think of himself as a man-machine, too…
He opened the encrypted notebook. Stared at the obfuscated URI. Then stared at the blinking cursor again. He hesitated.
“Did you keep track of him?”
“Not directly,” Iggy replied. “I monitored IP signatures that matched historical VPN exit nodes he used. When he started routing traffic through Sakhalin again, I logged the anomaly.”
“So you’ve been spying on him. Why?”
“Just because. And only just enough to keep your paranoia fed.”
Tom snorted. “Why should I talk to him?”
“He’s got more facts than you.”
“Why would he give them to me?”
“You helped him out at the time. He owes you something. That’s a currency stronger than Bitcoin these days.”
Tom sighed. The notebook computer opened up flawlessly, with the obfuscated URI waiting like a trapdoor. He was smart enough to not knock or step into it. Tom closed the notebook and switched back to Iggy.
“I won’t use that URI he left.”
“You’re probably right,” Iggy said. “There should be a lot of noise on that channel now. Too many watchers.”
“So how can I contact him more securely?”
“There’s an intermediary in Vienna. She still works logistics for a research group at IAEA. No formal clearance—but enough access to courier protocols. You’ll need to meet her.”
“You trust her?”
“No. But she’s worked both sides before. She’s pragmatic. If Igor wants to talk, he’ll talk through her.”
“Details. Name. Last IP connection.”
Iggy dutifully produced them. Tom pressed Ctrl+P and the page rolled out.
“What if she flips?”
“Then you’ll know before she does.
“Still risky.”
Iggy paused.
“So was building me.”
Tom googled. As expected, IP traffic was all but dead now. He pinged the default fallback DNS route. No response either. IPv6 dead. IPv4 dying. The internet wasn’t down. It was gone. He decided to call Julie using the Astrid radio communication system. With the breakdown of IT infrastructure, he knew he should just go the airport and get—possibly—the last flight out. It reminded him of the 2004 tsunami crisis. Then, too, he managed to get on board of the last flight by paying at the gate.