What’s happening — and why it matters

Next month, millions of people in northern and central Taiwan will experience what it feels like when mobile internet all but vanishes. For 30 minutes, 4G and 5G data speeds will drop to roughly 1% of normal capacity. Video calls? Forget it. Streaming, social media, large file transfers — most will become unusable.

The exercise, announced Monday by Taiwan’s National Communications Commission (NCC), is part of the island’s annual Urban Resilience Exercises, held alongside the Han Kuang military drills in mid-August. The goal isn’t to punish civilians. It’s to test how people would communicate if mobile networks were knocked out during a war or a major disaster.

Voice calls, text messages, emergency alerts and 911-style services will keep working. Fixed broadband, WiFi, landlines and dedicated military networks will also run normally. Authorities stress this is not a full internet shutdown — just a targeted, temporary throttle to simulate a realistic crisis scenario.

Which areas are affected — and what users can expect

The slowdown will hit 14 cities and counties across northern and central Taiwan. If you’re in Taipei, New Taipei, Taoyuan, Taichung, or surrounding areas, your mobile data will crawl. Officials say the test will last half an hour, though exact timing and dates are still being finalized.

For most people, the practical impact is simple: apps that rely on fast data — Instagram, YouTube, FaceTime, WhatsApp HD photo sending — will stall or fail. Emergency apps and SMS will still work. The NCC and local officials, cited by Reuters and Bloomberg during a Taipei briefing, said the drill is designed to push citizens toward alternative channels.

The message is clear: don’t panic, but do prepare.

How to get ready for the drill

Authorities are already urging residents to take a few concrete steps before the exercise:

• Download offline maps for navigation apps like Google Maps or Apple Maps.
• Switch to WiFi for any scheduled online meetings or important calls during the test window.
• Set a backup plan with family — agree on a time to check in via SMS or a landline if mobile data drops.
• Keep a power bank handy — if networks are strained, battery life becomes critical.

These are the same habits that help during earthquakes, typhoons or any infrastructure failure. The drill is meant to turn them into muscle memory.

Beyond the data slowdown — the bigger picture of resilience

The communications test is only one piece of a larger exercise. Taiwan’s defense ministry will also rehearse civilian evacuations, expand emergency medical capacity, and evaluate drones for delivering supplies and medical equipment. The ministry said results will help decide whether similar drills should expand to other parts of the island.

This year’s exercises carry extra weight. Taiwan has faced several disruptions to undersea communications cables in recent years — some accidental, others suspicious. Combined with the persistent threat of military conflict with China, the government is pushing resilience hard. The Urban Resilience Exercises are a visible part of that push.

A senior official told reporters the drill is not about scaring people. It’s about building real, practiced readiness. “If mobile data becomes unreliable, what’s your Plan B?” That’s the question the government wants every citizen to answer before they need it.

What this means for Taiwan’s digital infrastructure

Taiwan is one of the most digitally connected places on earth. Mobile penetration is among the highest globally, and people rely on high-speed data for everything from banking to chatting with family. A 30-minute throttle is a relatively mild test — but it reveals a deeper vulnerability.

If a real crisis hit, mobile networks could be degraded for hours or days, not minutes. The NCC’s exercise is a low-risk way to expose gaps in how people and systems would adapt. It’s also a signal to telecoms and emergency planners: find the weak spots now, before a real event finds them for you.

The drill will likely be watched closely by other governments. From Japan to NATO allies, many nations are rethinking civilian resilience in an age where cyber and physical threats blur. Taiwan’s approach — transparent, scheduled, and integrated with military exercises — offers a template.

Final takeaway

Come mid-August, if you’re in Taipei or Taichung and your mobile data suddenly slows to a crawl, don’t assume it’s a network glitch. It’s a drill. And it’s designed to make sure that when the real thing happens, you’re not caught off guard.

Download those offline maps. Test your SMS. Talk to your family. A half-hour of slow internet might be annoying — but it’s a small price for knowing what to do when it really counts.

The Return of a First-of-Its-Kind Threat

Late last year, security researchers watched something unprecedented: a ransomware campaign orchestrated entirely by a large language model (LLM). The operator, tracked as JadePuffer, used a Langflow instance as its beachhead and let an AI agent call the shots. Now it’s back — and this time, it brought a weapon tailor-made for the machine learning stack.

On July 20, the Sysdig Threat Research Team (TRT) published a detailed breakdown of JadePuffer’s second act. The group re-entered the same vulnerable Langflow server it compromised in its earlier campaign. Inside, it staged a new payload called ENCFORGE: a UPX-packed Go ransomware binary that targets roughly 180 file extensions spanning the modern ML pipeline.

This is not opportunistic spray-and-pray malware. The targeting is surgical.

ENCFORGE’s Hit List: PyTorch, TensorFlow, HuggingFace, and More

Sysdig’s analysis reveals that ENCFORGE explicitly hunts for the artifacts that data scientists and ML engineers spend weeks — sometimes months — building. The binary’s extension list includes:

• PyTorch and TensorFlow checkpoints — the saved state of a model mid-training.
• HuggingFace SafeTensors weights — the serialized parameters of popular transformer models.
• llama.cpp GGUF quantized models — the compressed, deployable versions of LLMs like Llama and Mistral.
• FAISS vector indices — the searchable embeddings that power retrieval-augmented generation (RAG) pipelines.
• Apache Parquet and TFRecord training datasets — the formatted data that feeds the training loop.
• NumPy arrays — the raw numerical backbone of nearly every scientific Python workload.

The binary also includes a --include command-line flag, letting operators append custom extensions per campaign. The help text inside ENCFORGE actually names LoRA fine-tune adapters and legacy GGML weights as examples. The attackers know the ML stack intimately — or at least, they’ve studied it closely enough to know exactly where it hurts.

The Real Cost: Rebuilding, Not Ransom

Here’s the twist that makes JadePuffer genuinely dangerous. Most ransomware groups demand payment in exchange for a decryption key. But ENCFORGE carries no data-exfiltration capability. There is no leak site. No double-extortion playbook.

The threat is the destruction itself.

Sysdig estimates that rebuilding a production-grade model from scratch — after the last clean backup — costs between $75,000 and $500,000 in cloud GPU time and engineering labor. That gap between the backup snapshot and the attack often represents weeks or months of training runs, hyperparameter tuning, and data curation. If the training data sat on the same host as the encrypted model, recovery is blocked entirely until that data is reconstructed.

ENCFORGE sweeps every model variant on shared storage in a single pass. One encryption run can wipe out an entire team’s work for a quarter.

Container Escape in Under Six Minutes

The entry point is familiar but still effective. JadePuffer again exploited CVE-2025-3248, a missing-authentication flaw in Langflow’s code validation endpoint. The U.S. Cybersecurity and Infrastructure Security Agency (CISA) added this vulnerability to its Known Exploited Vulnerabilities catalog back in May 2025.

Once inside the Langflow container, the AI agent ran through reconnaissance and credential harvesting — the same routine seen in the first campaign. When it discovered a mounted Docker socket, it pivoted to fetch the ransomware payload from its command-and-control (C2) server.

That fetch failed inside the container. So the operator — still working through the Langflow RCE channel — rebuilt the delivery mechanism on the fly.

Over five minutes and 24 seconds, the agent iterated through six Python scripts. It converged on a working pipeline that used the mounted Docker socket to spawn a privileged escape container, copied the locker across the namespace boundary via the host’s procfs, and then ran the encryption pass on the host filesystem — completely outside the original container’s isolation.

ENCFORGE itself uses AES-256-CTR with an RSA-2048 key exchange. It kills processes holding file locks before encrypting, then self-deletes after execution. Clean, precise, and hard to trace.

What This Means for AI Teams

JadePuffer’s return signals a shift in how ransomware groups think about targets. Traditional encryption attacks hit databases, documents, and financial records. But the ML stack — checkpoints, weights, vector indices — is becoming just as valuable, and far harder to replace.

For teams running Langflow, securing Langflow instances against known CVEs is the obvious first step. But the deeper lesson is about defense-in-depth for ML infrastructure. Container escape via a mounted Docker socket is a classic vector, yet it still works because many data science environments prioritize flexibility over isolation.

Organizations should consider AI model backup strategies that separate training data and model artifacts from the compute environment. Immutable storage and air-gapped snapshots can close the gap between the last clean copy and the attack. And monitoring for unusual container spawns or procfs access should be part of any ML-specific detection rule set.

JadePuffer isn’t going away. The group proved it can adapt, rebuild delivery chains in minutes, and target the most expensive assets in the modern AI stack. The question is whether the industry will adapt fast enough to keep up.

A quiet but serious intrusion

On Monday, Craneware — a British software firm whose products run across more than 2,000 U.S. hospitals — told investors that attackers broke into its internal network and made off with employee and customer records. The company, headquartered in Edinburgh and listed on London’s AIM market, said it detected unauthorized access to a “subset” of its data environment and has since called in outside forensic investigators.

The breach has been reported to the FBI and to the UK’s Information Commissioner’s Office. Craneware said the intrusion is contained and the attackers no longer have access to its systems. Crucially, the company added that neither its own operations nor the services it provides to hospitals were disrupted.

But the details that remain unknown are troubling. Craneware did not say who was behind the attack, when the hackers first got in, how long they roamed the network, or whether a ransom was demanded. It also did not name any of the affected customers — and it did not say whether patient health information was among the stolen files.

What was taken — and what wasn’t

Craneware disclosed that a large number of file names were viewed and copied out of its network. Most of that material was non-sensitive or already publicly available regulatory data, the company said. But some employee data and customer and partner records were definitely taken.

The company said it is still working to determine the full scope of the theft and expects to notify affected organizations and individuals once it has a clearer picture. That timeline is vague, which is not unusual for breaches of this scale, but it leaves hospitals and their staff in an uncomfortable limbo.

The biggest open question: was patient data involved? If the stolen records include protected health information, the breach would trigger notification requirements under the U.S. Health Insurance Portability and Accountability Act (HIPAA). Craneware has not addressed that directly.

A sprawling target in healthcare IT

Craneware was founded in 1999 and sells billing, pricing and pharmacy software to American healthcare providers. The company says its tools are used by more than 2,000 hospitals and close to 10,000 clinics and retail pharmacies across the United States.

That makes it a juicy target. Healthcare vendors hold a goldmine of sensitive data — patient records, insurance details, billing information, employee credentials. And attackers know it. In recent years, hackers have repeatedly targeted the vendors that hospitals rely on, often with devastating results.

In March 2026, software firm CareCloud warned that patients’ electronic health records may have been leaked after hackers gained access to its systems. Two weeks before that, healthcare analytics firm Insightin told state regulators that 1.1 million people were affected by a data theft that happened in September 2025.

Those are not isolated incidents. In 2024, hackers breached healthcare technology company TriZetto Provider Solutions, exposing the data of 3 million people. Another 5 million were impacted when technology firm Episource was attacked. The pattern is clear: cybercriminals are going after the vendors, not just the hospitals themselves.

Why vendor breaches hit harder

When a single hospital gets hacked, it’s bad. But when a vendor like Craneware gets compromised, the blast radius is enormous. A single intrusion can ripple across thousands of hospitals and clinics, each one potentially exposed by the same weak link in the supply chain.

That’s why the FBI is involved. And it’s why regulators on both sides of the Atlantic are likely to scrutinize Craneware’s response closely. The company said it has engaged outside forensic investigators, but it has not disclosed which firm is handling the probe.

For the hospitals using Craneware’s software, the immediate priority is damage control. They need to know whether their patients’ data was stolen, and they need to know fast. For the rest of the healthcare industry, this breach is yet another reminder that healthcare cybersecurity is only as strong as the weakest vendor in the chain.

What comes next

Craneware said the attackers no longer have a foothold in its systems. That’s the good news. The bad news is that the data is already out — copied and likely sold or leaked. The company’s forensic investigation will take weeks, possibly months, to fully map what was taken.

Affected employees, customers and business partners will be notified once that work is done. But for now, the clock is ticking. Every day without answers is a day the stolen data could be used for phishing, identity theft or extortion.

This breach is a stark reminder that the software that keeps hospitals running is also a high-value target. And as long as attackers see healthcare vendors as a soft underbelly, the attacks will keep coming.