Lazarus Group Linked to $6B Binance, $1.5B Bybit, $388M Bitget Hacks

According to data from the BIT Stock Platform, pre-market trading in US stocks saw a collective decline across the semiconductor & storage, crypto-related, and optical communication sectors. Intel’s (INTC) drop widened to 25%, driven by factors including its far weaker-than-expected Q4 guidance, an explosion at its Israeli factory, weak PC demand, and limited AI chip supply. Semiconductor and storage stocks: Seagate (STX) down 5.35%, Western Digital (WDC) down 5.94%, SanDisk (SNDK) down 4.27%, Micron (MU) down 4.17%, NVIDIA (NVDA) down 2.88%, Intel (INTC) down 25.49%, AMD (AMD) down 6.80%. Crypto-related stocks: MicroStrategy (MSTR) down 5.02%, Coinbase (COIN) down 4.66%, Circle (CRCL) down 4.66%, BitMine (BMNR) down 5.15%, SharpLink (SBET) down 5.95%. Optical communication stocks: FOTO (FOTO) down 6.53%, Marvell (MRVL) down 5.33%, Applied Optoelectronics (AAOI) down 7.97%, Lumentum (LITE) down 6.18%, Coherent (COHR) down 7.66%, Ciena (CIEN) down 4.68%, Fabrinet (FN) down 4.64%, AXT (AXTI) down 5.39%.

Whale Address Starting with 0x15a4 Liquidates 100+ Day BTC, ETH Long Positions, Net Profit ~$17.78M In the past hour, the whale with the address starting with 0x15a4 liquidated all its BTC and ETH long positions held for over 100 days, generating a total realized closing profit of approximately $20.554 million. After deducting funding fees and opening/closing trading fees incurred during the holding period for both positions, the net profit stood at around $17.779 million. The whale then flipped to short ETH and continued adding to its short positions, now holding 20,000 ETH shorts valued at roughly $44.614 million at an average price of $2,228.12 — twice the size of its prior ETH long position. No new BTC positions have been opened to date. This address previously entered positions at July lows, holding through the recent rally: - BTC: Opened a 1,000 BTC long on July 6 at an average price of $62,353.65, held for ~116 days before closing entirely at $78,181.69 average, delivering a closing profit of ~$15.828 million. - ETH: Initiated position building on July 8, expanded to 10,000 ETH on July 14, with an overall average entry price of $1,761.95. After ~114 days since first opening, it closed all ETH longs at $2,234.54 average, generating a closing profit of ~$4.726 million. During the holding period, ETH once rose to $2,809.5, ~59.5% above its cost; the exit price marked a ~26.8% gain from entry.

Beating AI News (Dongcha) – AI coding company Cognition has announced the acquisition of security startup Corridor, with the deal’s financial terms undisclosed. Corridor, which raised $25 million in funding in March this year at a $200 million valuation, specializes in addressing security risks from AI-generated code. Unlike traditional post-hoc vulnerability scanning, Corridor integrates directly with coding tools including Cursor, Claude Code, and Codex to inspect security risks as AI generates code, and can even block the merging of code with critical vulnerabilities. Its clients include LangChain, Lovable, and ElevenLabs. This acquisition has been in the works for some time. On September 29, Cognition hired Alex Stamos – former head of security at Corridor and ex-Facebook chief security officer – to lead its security operations. On October 26, it upgraded Devin Security, integrating cutting-edge security models such as Claude Mythos. The tool can not only scan for complex vulnerabilities but also assess their actual impact, enabling Devin to automatically reproduce, fix, and test issues. Corridor excels at blocking risks as code is being written, while Devin Security already has deep scanning and bulk remediation capabilities. Following the acquisition, Cognition plans to combine the two capabilities, enabling its AI to not only generate code but also oversee security throughout the development-to-deployment process.

Binance CEO He Yi has issued a security reminder: All information shall be subject to releases via official channels. Users are advised not to click non-official links or disclose their account information to anyone. Follow-up updates will be announced through official channels as soon as possible. Thank you again for your trust.

According to Yonhap News Agency, South Korea’s Financial Intelligence Unit (FIU) ordered the Korea Communications Commission on October 30 to block domestic access to two unregistered virtual asset operators, including terminating their services. On the same day, the Digital Asset Exchange Association (DAXA) referred three entities—Aegis Ventures, TetherMax, and SuperTrust—for investigation over alleged violations of the Special Financial Information Act and unregistered virtual asset business operations. The FIU ordered blocks on TetherMax and SuperTrust; Aegis Ventures has no independent website or mobile application, so it was not included in the block request. Under current law, unregistered virtual asset operators face a maximum penalty of five years in prison or a fine of 50 million won, and the law also applies to overseas operators serving South Korean residents. The FIU noted that unregistered operators operate in regulatory blind spots and may be used for money laundering. In June this year, a joint investigation by DAXA and compliant operators found that the average transaction fees of the identified unregistered operators ranged from 1.5% to 10%, up to 62 times the average level of South Korea’s five major won-based crypto exchanges (0.16%).

Beating AI News: Factory has released Droid 1.0, a coding agent runtime framework, introducing procedural tool calling (Code Mode) as its default execution method. The feature allows AI to first write JavaScript scripts, then sequentially call tools hundreds of times via the scripts, eliminating the need to re-request the large language model every few steps, thus reducing model call frequency and token consumption. For example, migrating and verifying 151 Linear tickets required 36 model calls and 3.1 million input tokens via the traditional approach; with Code Mode, only 2 calls and around 61,000 tokens were needed, a ~98% drop in input tokens. In broader tests, task costs for Claude Opus 5.5 saw a ~15% reduction. Factory also optimized cross-model tool calling strategies, caching, and task execution mechanisms. In its official Terminal-Bench 4 self-test, using the same GPT-6.1 Sol model, Droid scored 3.5 points higher than Codex, with a 12% lower successful task cost, though execution time increased by 10%. It’s worth noting that Code Mode is not a new technology: Cloudflare and Anthropic previously unveiled similar solutions, while Pi, OpenCode 2.0, and DeepSeek Harness also support the feature.