Cisco Cuts 4,000 Jobs in Major Shift Toward AI Investment

Cisco Systems reported "record" revenue in its fiscal 2026 third-quarter earnings, while also revealing plans to cut approximately 4,000 jobs. CEO Chuck Robbins explained that despite achieving Record Quarterly Revenue, the company needs to restructure its organization to accelerate investment in AI.

This move symbolizes how the transition from the traditional networking equipment business to AI infrastructure is affecting corporate employment structures. Cisco has already carried out multiple rounds of layoffs in recent years, and this is the latest. While expanding AI investment is an industry-wide trend, the pattern of "layoffs even amid record revenue" underscores just how fierce the AI competition has become.

10 Chinese Companies Granted US AI Chip Export Licenses, but Beijing Blocks the Purchases

It has been reported that the US Commerce Department approved purchases of up to 75,000 Nvidia H200 chips by about 10 Chinese companies, including ByteDance, Alibaba, and Tencent. However, according to Commerce Secretary Lutnick, the Chinese mainland government has blocked these purchases, and not a single chip has actually been shipped.

Beijing's aim appears to be protecting its domestic AI industry. China is promoting the development of its own AI semiconductors, such as Huawei's Ascend chips, and has laid out a policy of reducing dependence on US products. This odd situation — export licenses granted, but nobody buying — signals that the US-China tech cold war has entered a new stage.

Alibaba's Qwen-Image-2.0 Dramatically Improves Image Generation Efficiency

The technical report for Qwen-Image-2.0, released by Alibaba, is attracting attention. The model achieves roughly twice the compression rate of its competitors while cutting generation steps from the previous 40 down to just 4 — effectively a 10x speedup in generation.

On the technical side, the key points include improvements to training stabilization techniques and an architecture design that resolves the compression-quality trade-off more favorably than before. Computational cost is a major challenge for image generation AI, and Qwen-Image-2.0's approach could also influence its practicality on edge devices.

Claude AI Recovers $400,000 from an 11-Year-Old Bitcoin Wallet

A case is making headlines in which roughly $400,000 (about 60 million yen) was recovered from a Bitcoin wallet whose password was forgotten 11 years ago, using Claude AI. According to Tom's Hardware, the wallet's owner leveraged Claude to automate the password-guessing process and, after 3.5 trillion attempts, succeeded in decrypting an old wallet backup.

It is drawing attention as one of the first major cases in which AI was used for cryptocurrency recovery at a practical level. Note, however, that this method assumes a wallet backup exists, so it cannot be applied to every lost wallet.

AI Finds Critical Linux Kernel Vulnerabilities Three Weeks in a Row

According to ZDNet, an AI tool has discovered its third critical Linux kernel vulnerability in two weeks. This streak of discoveries demonstrates that AI-powered security audits can efficiently identify flaws that traditional manual reviews tend to miss.

AI-driven vulnerability discovery is becoming a turning point in open source software maintenance. In massive codebases like the Linux kernel, the limits of human reviewers have become apparent, and AI's complementary role is increasingly likely to contribute to substantive security improvements.

MIT Develops RLCR, a New Method That Makes AI Say "I Don't Know"

MIT's CSAIL (Computer Science and Artificial Intelligence Laboratory) has announced RLCR (Reinforcement Learning from Calibration Rewards), a new method that makes AI models output "I don't know" for answers they lack confidence in.

Current reasoning models tend to respond with the same degree of confidence whether an answer is accurate or a guess, and this "overconfidence" leads to reduced reliability in real-world operations. RLCR applies specific modifications to the model's training process, enabling it to express uncertainty appropriately. As a fundamental effort to improve AI reliability, it is expected to find a wide range of applications.

---