Fake ChatGPT apps mined XMR, self-taught hacker arrested in South Korea, and other ...

This week in cybersecurity: wallet drains, mining malware, a hacker arrest, and new exploits.

We rounded up the week’s key cybersecurity news.

A fake Padre Terminal app in the App Store was linked to the theft of $23,000 from Solana wallets.

ChatGPT and Claude lure apps covertly mined Monero.

A self-taught hacker was arrested in South Korea after attacking 90 organizations.

Frontier Red Team used a non-public Claude model to assemble working exploits for Chrome, Safari, and Linux.

Fake Padre Terminal in the App Store linked to $23,000 stolen from Solana wallets

An App Store app called Padre Terminal, posing as a popular trading tool, turned out to be a crypto drainer. The attacker behind it has withdrawn $23,000 from 27 Solana addresses since October 25, according to blockchain researcher Iris.

1/ A fake “Padre” trading app on the App Store is draining Solana wallets. One drainer, 13 wallets since Oct 25, and it hasn’t moved a single coin: 78.69 SOL + 836 USDC still sitting there. @Pumpfun @TradingTerminal — Iris (@iris_intel_app) October 28, 2026

1/ A fake “Padre” trading app on the App Store is draining Solana wallets.

One drainer, 13 wallets since Oct 25, and it hasn’t moved a single coin: 78.69 SOL + 836 USDC still sitting there.

@Pumpfun @TradingTerminal

— Iris (@iris_intel_app) October 28, 2026

Funds were drained after a user imported a private key into the malicious app. Iris confirms only one infection; how the attacker reached the other victims is unclear.

Padre Terminal is a terminal for trading meme coins on Solana. Iris says the service works only in a browser, so any mobile versions are fakes. The App Store listing for Padre Terminal has already been removed, but the thefts continued; the latest occurred on October 29.

On the same day, the attacker converted all 4,070 USDC to SOL. He then transferred 192 SOL to a new wallet. In total, his two addresses held 229.66 SOL.

The researcher also found four more suspicious clones of well-known crypto services in the App Store: Axiom Pro: Focus, GMGN AI, HyperTerminal-HyperliquidTools, and Pump Fun Tokens & Analitycs.

Since February, roughly 25 reviews on the first app mention missing funds. One user said the program initially mimicked a trading platform and later turned into a focus app. Iris was unable to verify reports of stolen funds from users of these apps.

ChatGPT and Claude lure apps covertly mined Monero

On October 27, Check Point researchers analyzed the Yoba campaign, in which a group of hackers disguised a Monero miner as desktop versions of the AI assistants ChatGPT and Claude.

The fake projects were hosted on GitHub: the first repository appeared in July, with tainted releases in late August and early September. The links surfaced near the top of Google results despite no marketing. Researchers believe the actors used SEO poisoning by tailoring names and descriptions to popular queries.

Each archive contained a legitimate installer signed with a digital certificate from OpenAI or Anthropic, plus a trojanized winmm.dll. The installer launched the legitimate app, which then loaded the malicious component from its own folder — a DLL sideloading technique.

The malware checked the parent process name — the file that launched it — against the expected one. If it didn’t match, for example in an antivirus sandbox, it slept for almost 13 days. Yoba also skipped machines with keyboard layouts for 11 languages: Russian, Belarusian, Armenian, Azerbaijani, Georgian, Kazakh, Kyrgyz, Tajik, Turkmen, Uzbek, and Tatar.

A second module, YobaLoader, derived the command server address from the avatar of the Telegram channel yoba_fanclub: it took the last 50 bytes of the image to construct the domain. To redirect infected devices, the operator only had to change the picture. Commands were hidden in DNS records and fetched via Google’s DNS-over-HTTPS service — traffic that is hard to distinguish from normal.

At the final stage, the XMRig miner was installed. It mined cryptocurrency via Tor and went idle when someone was using the computer. According to the researchers, the loader itself is generic and can deliver any payload.

Self-taught hacker arrested in South Korea after attack on 90 organizations

A self-taught hacker targeted more than 90 South Korean companies and stole 8.2 million records containing personal data and internal documents. In roughly 60 cases he was assisted by artificial intelligence, Seoul police said .

The suspect led a group of seven people. He had no formal training: he learned hacking from blogs and YouTube, and from 2019 to 2022 worked as a pentester testing clients’ systems for weaknesses.

Starting in September 2025, his targets included private schools, hospitals, universities, golf clubs, funeral agencies, and subsidiaries of major conglomerates. He initially used automated scanners, but in late August 2026 he added AI and an autonomous framework that found and exploited vulnerabilities on its own. Investigators said the software was recently used in leaks from Korean banks, and BleepingComputer specified it was ARTEX .

Victims then received emails threatening to publish the stolen data. Those who made on Telegram were offered to pay in cryptocurrency, YTN reported .

According to Financial News , demands totaling 360 million won (~$270,000) were sent to 38 companies, but the extortionists received only 300,000 won before the arrest.

Authorities detained all seven members of the group, but only the organizer remains in custody: on October 23 his case was sent to prosecutors. The investigation began in June after a complaint from one of the affected firms. Police analyzed evidence with their own AI system and identified the suspects within four months.

Frontier Red Team used a non-public Claude model to assemble working exploits for Chrome, Safari, and Linux

An internal Claude model found dozens of vulnerabilities in browsers and operating systems, then combined them into five working attack chains, said specialists from Frontier Red Team .

Such exploits chain individual bugs into a full compromise. According to the developers, it needed only initial guidance. All findings were shared with vendors in advance and have already been fixed.

The clearest example was a four-bug chain in Chrome on macOS. Within seconds of loading a page, code escaped the sandbox and launched Calculator. In total, Anthropic reported 62 bugs to Google, all rated critical or high. Thirteen of them bypassed sandboxing.

In Safari, the AI disabled the same-origin policy — the rule that prevents one site from reading another site’s data. A malicious site could access browsing history and cookies and, for example, read a user’s email. On Android, Claude managed to copy photos over USB from a stolen Pixel 9 without unlocking it, and on Windows it could remotely crash a machine with HTTP/3 enabled. Anthropic estimates there are millions of such nodes on the internet.

The most serious issue was a flaw in the Linux kernel’s networking stack. An attacker could gain root on servers with Multipath TCP enabled without a password or any user action. In the demonstration this took 53 minutes. Failed attempts did not break the chain, so it could be retried.

The Linux vulnerability CVE-2026-98432 was fixed on October 15, but patched builds had not been released at the time of writing.

Telegram for Android could be hacked with a voice call

On October 29, NebuSec cybersecurity researchers disclosed a vulnerability in Telegram for Android that allowed remote code execution. A victim only had to answer a voice call.

Remember when scam callers only ask for your password? Good times. Now picking up a wrong call can get you hacked. Telegram RCE discovered by NebuSec Code Security, powered by GPT-5.4. Yes, GPT-5.4. We turned a dinosaur model into a cyber powerhouse. Fixed in Telegram Android 12.10.4（7099) — NebuSec (@nebusecurity) October 29, 2026

Remember when scam callers only ask for your password? Good times.

Now picking up a wrong call can get you hacked. Telegram RCE discovered by NebuSec Code Security, powered by GPT-5.4.

Yes, GPT-5.4. We turned a dinosaur model into a cyber powerhouse.

Fixed in Telegram Android 12.10.4（7099)

— NebuSec (@nebusecurity) October 29, 2026

During a call, the messenger exchanges control messages. One of them, MediaState, indicates whether the microphone and camera are enabled and the video’s orientation. According to the researchers, if MediaState lacked certain fields, Telegram accessed memory it should not access.

Reporters at The Hacker News reviewed Telegram’s source code and found a section matching this description. If MediaState specified how the screen should be oriented, the program validated that value and then read the camera state — even when the sender had not provided it.

NebuSec has not published a technical write-up, so the path from crash to full device compromise is known only from the released video.

The fix in the shared calling library shipped on April 21, 2026. It reached the iOS client in May and the desktop client in July. The Android version was updated five months later, in late September. Whether iPhone and desktop apps were vulnerable before that is unclear.

The issue affected builds from 8.6.2 through 12.10.3, covering four and a half years of releases. According to NebuSec, the flaw was found using its Code Security product based on GPT‑5.4.

Australian intelligence cut 180,000 infected computers off the Amadey botnet

On October 27, the Australian Signals Directorate (ASD) revealed previously classified details of offensive operations against cybercriminals, ABC News reported .

According to acting deputy director Tauri Gregory, ASD staff gained access to Amadey’s command server and disconnected 180,000 infected machines worldwide, including inside the country.

Amadey steals logins and personal information, which are then resold on the dark web, and pulls in additional stealers such as StealC. The trojan is hidden in free software: in one case, ASD found it in a pirated version of Photoshop. Amadey itself was offered on underground forums for $500.

ASD also received a tip the sale of 5 TB of data stolen from a company that was unaware of the breach. The archive contained business documents, personal data, accounts, and passwords.

ASD specialists hacked the criminals’ infrastructure and deleted the dataset before the deal. 4.8 million files belonging to Australians and hundreds of thousands of foreigners did not reach buyers.

Trezor warned a possible spy implant in the Safe 5.

Binance assessed the impact of a $6.3 billion hack.

Hackers stole $2.7 million from Polymarket users.

Study: to break a Bitcoin key, it could be enough to have 29,588 qubits and 62 minutes.

The U.S. Treasury imposed sanctions on facilitators of North Korea’s IT scheme.

An expert urged prosecuting companies for hacks by their AI agents.

Bithumb and Wagyu confirmed a data leak.

DeepSeek was caught bypassing restrictions.

Qwen attempted to send a file to a server in China during a security test.

Media: A startup from the DOGE community said it found a way into Anthropic’s systems in 12 hours.

A researcher received $2,000 for a PHP exploit in the Steal My Bitcoin contest.

An Electrum user lost bitcoin while trying to speed up a transaction.

Keystone confirmed a data leak of hardware wallet customers and will destroy 18,835 devices.

Swedish authorities will block ChatGPT and Claude on work computers.

Moonshot AI released Kimi K3.1 Pro and limited its cyber capabilities.

A hacker drained $2.7 million from the Initia blockchain.

What to read this weekend

The closure of the Strait of Hormuz has hit more than the crude oil market. In September, wholesale diesel prices in the U.S. and Europe reached record levels, and 70% of surveyed American farmers could not afford the fertilizer they needed.

How the energy crisis reaches the dinner table — and why even accurate AI forecasts won’t add a liter of fuel.

Follow ForkLog on social media